Photovoltaic Systems Advances in Research and Applications 9798886979077

Citation preview

Renewable Energy: Research, Development and Policies Energy Science, Engineering and Technology

No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in rendering legal, medical or any other professional services.

Renewable Energy: Research, Development and Policies Solar Collectors and Systems Mamdouh El Haj Assad, DSc (Editor) Mohammad Alhuyi Nazari, PhD (Editor) 2023. ISBN: 979-8-88697-774-5 (Softcover) 2023. ISBN: 979-8-88697-860-5 (eBook) The Future of Solar Power Professor Dr. Hussain H. Al-Kayiem (Editor) 2023. ISBN: 979-8-88697-709-7 (eBook) More information about this series can be found at https://novapublishers.com/product-category/series/renewable-energy-researchdevelopment-and-policies/

Energy Science, Engineering and Technology Fuel Briquettes Made of Carbon-Containing Technogenic Raw Materials Nina Buravchuk, PhD (Editor) Olga Guryanova (Editor) 2023. ISBN: 979-8-88697-907-7 (Softcover) 2023. ISBN: 979-8-88697-944-2 (eBook) The Fundamentals of Thermal Analysis Mamdouh El Haj Assad, PhD (Editor) Ali Khosravi, PhD (Editor) Mehran Hashemian, PhD (Editor) 2033. ISBN: 979-8-88697-759-2 (Hardcover) 2033. ISBN: 979-8-88697-875-9 (eBook) More information about this series can be found at https://novapublishers.com/product-category/series/energy-science-engineeringand-technology/

Sudip Mandal and Pijush Dutta Editors

Photovoltaic Systems Advances in Research and Applications

Copyright © 2023 by Nova Science Publishers, Inc. All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or otherwise without the written permission of the Publisher. We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to reuse content from this publication. Please visit copyright.com and search by Title, ISBN, or ISSN. For further questions about using the service on copyright.com, please contact: Copyright Clearance Center Fax: +1-(978) 750-4470

Phone: +1-(978) 750-8400

E-mail: [email protected]

NOTICE TO THE READER The Publisher has taken reasonable care in the preparation of this book but makes no expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is assumed for incidental or consequential damages in connection with or arising out of information contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary damages resulting, in whole or in part, from the readers’ use of, or reliance upon, this material. Any parts of this book based on government reports are so indicated and copyright is claimed for those parts to the extent applicable to compilations of such works. Independent verification should be sought for any data, advice or recommendations contained in this book. In addition, no responsibility is assumed by the Publisher for any injury and/or damage to persons or property arising from any methods, products, instructions, ideas or otherwise contained in this publication. This publication is designed to provide accurate and authoritative information with regards to the subject matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering legal or any other professional services. If legal or any other expert assistance is required, the services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.

Library of Congress Cataloging-in-Publication Data ISBN: 979-8-89113-102-6

Published by Nova Science Publishers, Inc. † New York

Contents

Preface

.......................................................................................... vii

Chapter 1

Advances in the Artificial Intelligence Models for Photovoltaic Systems ...................................................1 Ekaterina A. Engel

Chapter 2

Recent Advances in Photovoltaic Materials and Technology ................................................................63 Sukla Basu

Chapter 3

Carbon Nanodots in Photovoltaic Cells, a Solar Energy Harvester: A Critical Review ...............75 Biswajit Gayen, Pijush Dutta and Chayan Goswami

Chapter 4

Power Converters for Efficient Control and the Utilization of Solar Photovoltaic Energy .......................95 Shib Sankar Saha and Biswamoy Pal

Chapter 5

A Particle Swarm Optimization Approach for the Maximum Power Point Tracking of a Grid Connected Shaded Photovoltaic Generator ................125 Mouna Ben Smida and Anis Sakly

Chapter 6

An FDTD Study on the Broadband Light Absorption Enhancement in Thin Film Solar Cells Using Metal Nanoparticle Arrays .......................149 Saritha K Nair and V. K. Shinoj

Chapter 7

Photovoltaic Systems and Applications for the Early Diagnosis of Skin Diseases in Humans Using Artificial Intelligence ...........................159 Paryati

vi

Chapter 8

Contents

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels in Dairy Farms with Different Fixed Tilt Angles ...........................................183 Antonio José Steidle Neto and Daniela de Carvalho Lopes

About the Editors ......................................................................................203 Index

.........................................................................................205

Preface

Renewable energy sources play a vital role in the production of electrical energy as they overcome dependency on non-renewable resources. Photovoltaic (PV) System is very popular and wide-used renewable energy that uses solar energy for electricity production worldwide. At present, the PV market as one of the fastest growing industries. Many researches are going on to increase the efficiency of solar cell. Moreover, latest trends like machine learning and IoTs are also emerged together to improve performance of the system further. This Book aims to provide the latest research and development on PV system as well as provide some critical review that will help to increase the knowledge about the PV system to the readers. Some reviews have been given to the different Photovoltaic Materials and Technology, Carbon Nanodots. On the others hand, implementation of Artificial Intelligence and Machine Learning for modeling of PV system and application of optimization technique (such as Particle swarm Optimization etc.) for solar tracking to improve the efficiency of the solar array has also provided. Use of nano-particle for broadband light absorption enhancement by solar cell is another futuristic approach. How PV system can be used for AI based skin disease diagnosis is also proposed by the authors. Hope, the readers or research community will be benefitted a lot who are interested in the field of PV System. I must thankful to all the authors without their valuable contributions towards the Book, the timely publication of this Edited Book “Photovoltaic Systems: Advances in Research and Applications” is not possible. I must appreciate all the co-editors and reviewers for their valuable comments and suggestions to complete this journey. Last not the least, I am also grateful to Nova science Publishers to give their opportunity to be the Editor of this Book.

Chapter 1

Advances in the Artificial Intelligence Models for Photovoltaic Systems Ekaterina A. Engel ∗

Department of Information Technologies and Systems, Katanov State University of Khakassia, Abakan, Russian Federation.

Abstract The real-life photovoltaic systems have complex nonlinear dynamics with uncertainties A PV system has complex nonlinear dynamics with uncertainties due to variations in system parameters and insolation. Thereby, it is difficult to approximate these complex dynamics with conventional algorithms whereas Artificial Intelligence (AI) methods yield the essential performance required. AI models are key units in recent PV systems for design, forecasting, maintenance, and control to provide the best safety, reliability, robustness, and performance as compared to classical algorithms, which are usually employed in the hardware and software of PV systems. Considering this, the goal of our study is to explore and analyze AI models and their advantages and shortcomings as compared to classical algorithms for the design, forecasting, maintenance, and control of PV systems. In contrast with other review, our research briefly summarizes our intelligent, selfadaptive models for sizing, forecasting, maintenance, and control of a PV system; sets benchmarks for performance comparison of the reviewed AI models for a PV system’s system; proposes a simple but effective integration scheme of an AIPV system’s implementation and outlines its future digital transformation into a smart PV system based on the integrated cutting-edge technologies; and estimates the impact of AI models based on the proposed scheme on a PV system value chain. ∗

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

2

Ekaterina A. Engel

Keywords: AI, ML; neural networks; DL; PV; PV system; smart model

Abbreviations SCADA Supervisory control and data acquisition ML Machine learning PID Proportional integral derivative CRISP DM Cross Industry Standard Process for Data Mining ONNX Open neural network exchange PCA Principal Component analysis NN Neural network LR Linear regression SVM Support vector machine RF Random Forest DT Decision tree DL/DNN Deep neural learning/network ANN Artificial neural network RNN Recurrent neural networks CNN Convolutional neural networks XGBoost Extreme gradient boosting SGD Stochastic gradient descent ACC Accuracy MCC Matthew’s correlation coefficient ROC Receiver operating characteristic AUC Area under the curve MAE Mean absolute error CEEMDAN Complete ensemble empirical mode decomposition with adaptive noise nRMSE Normalized RMSE nMAE Normalized MAE STLBO Simplified Teaching Learning Based optimization ANFIS Adaptive network based fuzzy inference system PSO Particle swarm optimization QK-CNN quad-kernel deep CNN MPPT Maximum power point tracking NSRDB National solar radiation data base MPP Maximum power point CGO Chaos game optimizer

Advances in the Artificial Intelligence Models for Photovoltaic Systems

3

CARO Applied chaotic reproduction optimization RH Relative humidity CSA Cuckoo search algorithm BMO Bird Mating Optimization algorithm MSSO Modified simplified swarm optimization algorithm CI Cloud index WS Wind speed Pr pressure MD QPSO Multidimensional quantum behaved particle swarm optimization C-LSTM Constrained LSTM kNN k-Nearest Neighbors ETR Extra trees regressor RMSE Root-mean square error CWT Continuous wavelet transform IFR Infrared UAV Unmanned Aerial Vehicle IS Isolation Forest LOF Local Outlier Factor STC Standard test conditions DQL Deep Q-learning DDPG deep deterministic policy gradient RBF NN radial basis function neural network TCT Total-cross-tied SD-PAR Shade Dispersion Physical Array Relocation MHHO Modified Harris Hawks Optimizer BPFPA Bee Pollinated Flower Pollination Algorithm FPA Flower Pollination Algorithm MPCOA Mutative Scale Parallel Chaos Optimization Algorithm MFNN Modified fuzzy neural net R-JADE Repaired adaptive differential evolution TVIWAC PSO PSO with time varying inertia weight and acceleration coefficients BBO-HCS Biogeography optimization algorithm based heterogeneous cuckoo search ABC Artificial Bee Colony NMS Nelder Mead algorithm BBO Biogeography Based Optimization LM Levenberg-Marquardt

4

Ekaterina A. Engel

PSA IADE GGHS ABSO IGHS HS SA

Parallel Swarm Algorithm Improved Adaptive Differential Evolution Grouping based global harmony search Artificial Bee Swarm Optimization Innovative Global Harmony Search Harmony Search Simulated Annealing

Nomenclature 𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡) �𝑥𝑥 (𝑡𝑡) 𝑑𝑑 𝑑𝑑 (𝑡𝑡)

𝑥𝑥𝑋𝑋,𝑗𝑗𝑋𝑋

Сurrent encoded dimension of X position Personal best encoded dimension of position X j-th component of the particle X position

𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑗𝑗𝑑𝑑 (𝑡𝑡)

j-th component of global best swarm position in

𝐷𝐷�𝑑𝑑ℎ,𝑞𝑞 �

Dimension of particle X Dimension component of particle X Estimate of Lipschitz constant of an RNN Maximum number of hidden units 𝐻𝐻 Fitness function based on the Chebyshev criterion Total rate of insolation Air mass Altitude angle Solar azimuth angle PV module azimuth angle Reflection factor PV module tilt angle Sky diffuse factor

encoded dimension d g(d) 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝐽𝐽𝐼𝐼 𝜇𝜇 𝛽𝛽 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 𝛬𝛬𝐷𝐷 𝑑𝑑ℎ𝑗𝑗 𝑑𝑑

𝛬𝛬ℎ 𝐻𝐻 𝑓𝑓(𝑋𝑋) GC m β φS φC p 𝛴𝛴 C

Index of the global best swarm’ particle Global best swarm’ encoded dimension Jacobian matrix, Learning rate Contraction–expansion coefficient Optimum architecture of the MFFN Estimate of Lipschitz constantof the dataset Hidden neurons’ number

Advances in the Artificial Intelligence Models for Photovoltaic Systems

A, k G0ht Pht Ph t-2 Cht-2 lht Rht Wht dht aht 𝜇𝜇𝑗𝑗 (𝑠𝑠) 𝐹𝐹𝑗𝑗 (𝜇𝜇𝑗𝑗 (𝑠𝑠), 𝑠𝑠) 𝑌𝑌�𝑠𝑠 𝑖𝑖 � f 𝐼𝐼 𝑉𝑉 Ir 𝑇𝑇 𝑖𝑖 𝑡𝑡 𝑖𝑖 𝑃𝑃𝑖𝑖 𝐼𝐼𝑚𝑚𝑖𝑖 𝑥𝑥 𝑖𝑖 𝑢𝑢𝑖𝑖

5

Parameters related to the Julian day number Extraterrestrial insolation Power from a PV array Historical data of the power from a PV array Historical data of clear-sky index Cloudiness (%) Pressure Wind speed Wind direction Ambient temperature Membership function Tuned RNN Tuned two-layered RNN Time of the last PV box’s fault DC current Voltage Irradiance Ambient temperature Panel temperature Wiring losses’ F-FOP of a PV box Image of PV modules Input signal of MFNN Output signal of MFNN

Introduction PV systems have complex nonlinear dynamics with uncertainties since the system's parameters and insolation fluctuate (Kurukuru et al., 2021). Thereby, it is complicated to approximate these complex dynamics with classical algorithms, while AI methods provide the required performance (Forootan et al., 2022). In modern PV systems, AI models are crucial units to increase the quality of big dataset processing for PV system design, forecasting, maintenance, and control (Kurukuru et al., 2021; Forootan et al., 2022). Within the EU COVID-19 strategic reply, the smart energy standards define a cloud platform specification for a distributed solar big data ecosystem that will provide the creation of effective AI models for smart solar energy (EITCI Institute). Within breakthrough studies, AI models collected, analyzed, and

6

Ekaterina A. Engel

converted a huge number of datasets into AI knowledge because of traditional algorithms’ ineffectiveness. These big data sets are collected by supervisory control and data acquisition (SCADA) systems (Massaro et al., 2022). The contribution of this study is threefold. First, we reviewed more than 100 research papers devoted to state-of-the-art AI models of PV systems; most of the articles were published in the last five years. Second, we reviewed resources where researchers can find open datasets, source code, and AI framework and simulation environments to create AI models for a PV system. Third, in contrast with other review, our study proposes a simple but effective pipeline scheme for an AIPV system’s implementation and outlines its future digital transformation into a smart PV system based on integrated, cuttingedge technologies; estimates the impact of the AI models based on the proposed scheme on a PV system value chain; sets benchmarks for performance comparison of the reviewed AI PV system models; and briefly summarizes our self-adaptive models for sizing, forecasting, maintenance, and control of a PV system based on a modified fuzzy neural net (MFNN) (Engel, 2020; Engel et al., 2020; Engel et al., 2018, Engel, 2021).

AI Models for a PV System Real-life PV systems have complex, nonlinear dynamics due to variations in system parameters and insolation. Thus, AI methods have been proposed to approximate this complex dynamic. The recent studies (Kurukuru et al., 2021; Forootan et al., 2022; Engel, 2020; Engel et al., 2020; Engel et al., 2018; Engel, 2021; Engel et al., 2020; Gaviria et al., 2022; Youssef et al., 2017; Engel et al., 2022; Berghout et al., 2021) prove that AI models for a PV system’s design, forecasting, maintenance, and control increase the effectiveness and reliability as compared to conventional algorithms. In smart PV systems, AI models are crucial units to increase the quality of datasets processing for the PV system’s design, forecasting, maintenance, and control. Big data from PV systems combined with weather big data, enables the creation of AI models to solve complex tasks of a PV system’s design, forecasting, maintenance, and control.

AI model The reliability, accuracy, and other demanded quality parameters must be composed as the performance of an AI model. This model must be created

Advances in the Artificial Intelligence Models for Photovoltaic Systems

7

effectively with high-quality datasets to have optimal performance (Ha et al., 2020). Figure 1 show the basic life cycle of an ML model.

Figure 1. A basic life cycle of an ML model.

ML model creation has two phases: data preparation (DP phase) and model creation (MC phase). They should be elaborated by the Cross-Industry Standard Process for Data Mining cycle (CRISP-DM) (Nguyen et al., 2019) and Open Neural Network Ex-change (ONNX) format (https://onnx.ai). The datasets are preprocessed (Figure 1) in a simple way (standardization or encoding). Data preparation algorithms include dimensionality reduction (principal component analysis (PCA)), sampling (subsampling, oversampling), transformation, encoding, feature extraction, and selection (Ha et al., 2020). Feature extraction is a crucial step in a smart PV system’s creation because it provides knowledge for ML model creation (Ha et al., 2020). The DM algorithms generate features. The most relevant data are further separated into train, validating, and test datasets (Figure 1). An ML model to solve either classification or regression tasks are trained based on a train dataset. The ML frameworks provide an automatic MC phase, including validating (Figure 1). The trained ML model is deployed. If a monitored ML model does not provide optimal performance, then it is retrained based on updated datasets.

8

Ekaterina A. Engel

ML Methods for a Smart PV Model Creation An ML model can be developed based on neural network (NN) or non-NN algorithms (Ha et al., 2020). The last ones include PCA, Random Forest (RF), support vector machine (SVM), and Decision Tree (DT). In contrast with nonNN algorithms, NN architectures can include various neurons, which are specified by ONNX (Khan et al., 2021). A deep neural learning/network (DL/DNN), such as a recurrent neural network (RNN), convolutional neural network (CNN), and transformers, is part of the ML methods with feature learning that use complex connectivity architectures to automatically mine meta features from the input. NNs, such as artificial neural networks (ANNs), radial basis function neural networks (RBF-NNs), generative adversarial networks (GANs), RNNs, and CNNs have recently made major progress in practical applications of solar energy (Kurukuru et al., 2021). Figure 2 shows two NN methods’ classes and the ML method groups according to the task they solved for a PV system (Forootan et al., 2022). The ensemble’s types are bagging, boosting, and stacking/blending (Brown et al., 2017; Shahhosseini et al., 2022). Table 1 presents the comparison of ensemble techniques (Brown et al., 2017). There are constant and dynamic weighting ensemble approaches. In recent studies, the most used ensemble methods are RF, Extreme Gradient Boosting (XGBoost), Extreme Learning Machine (ELM), etc. Model training algorithms that optimize performance include quasiNewton, stochastic gradient descent (SGD), evolutionary computation, genetic programming, etc. (Nguye et al., 2019). The creation of the ML model is the most complex and important task which includes the creation of an optimal ML model’s architecture and requires a multidimensional global optimization (GO). A lot of techniques provide a model’s evaluation, including crossvalidation, kfold, holdout with a different performance including accuracy (ACC), mean squared error (MSE), precision, receiver operating characteristics (ROC), recall, Matthew’s correlation coefficient (MCC), F1, area under the curve (AUC), mean absolute error (MAE), and root-MSE (RMSE). The relative errors, such as normalized RMSE (nRMSE), normalized MAE (nMAE), etc., facilitate the comparison between models that are tuned based on datasets with different scales.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

9

Figure 2. A Classification of tasks that are solved based on ML methods.

Table 1. Comparison of ensemble techniques Name of Method Bagging

Advantages Tends to reduce variance more than bias

Boosting

Reduces bias and variance

Stacking/blending

Provides the optimal combination of base learners, reduces variance, and bias [18]

Disadvantages Does not work well with relatively simple models Sensory to noise and outliers in data. Susceptible of overfitting In the case of huge datasets, the computational time increases sufficiently as each classifier is working independently on the huge dataset.

With the goal to develop intelligent models for sizing, forecasting, and control of a solar plant system and to make an RNN more adaptive with regard to a task’s complexity and overfitting problem, we developed an MFNN (Engel, 2020; Engel et al., 2020; Engel et al., 2018; Engel, 2021). The MFNN includes RNNs with fuzzy units and/or a convolutional block to process images. An RNN approximates a membership function in contrast to an Adaptive Network-Based Fuzzy Inference System (ANFIS). We combined the modified multidimensional quantum-behaved particle swarm optimization

10

Ekaterina A. Engel

(PSO) with the Levenberg–Marquardt algorithm (MD QPSO) and developed a hierarchical encoder of the particle’s dimension component (Engel, 2020; Engel et al., 2020; Engel et al., 2018; Engel, 2021) to automatically create an optimal architecture of an MFNN and improve the convergence.

The modified MD QPSO We elaborate the modified MD QPSO (the termination criteria are {𝑇𝑇, 𝜀𝜀𝐶𝐶 , .  . . }; 𝑆𝑆 is the number of particles) as follows: Step 1. For ∀𝑋𝑋 ∈ {1, 𝑆𝑆}do: Generate 𝑑𝑑𝑋𝑋,𝑞𝑞 (1) = 𝑎𝑎𝑎𝑎𝑎𝑎(𝑁𝑁(0,1)),𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡) = 𝑎𝑎𝑎𝑎𝑎𝑎(𝑁𝑁(0,1)) – current ����� encoded dimension of X position, 𝑞𝑞 = 1. .3. �𝑥𝑥 (𝑡𝑡)–personal best encoded dimension of Initialize 𝑑𝑑̃𝑋𝑋 (0) = 𝑑𝑑𝑋𝑋 (1), 𝑑𝑑 position X. For ∀𝑑𝑑 ∈ {𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 , 𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 } do: 𝑑𝑑 (𝑡𝑡) Generate 𝑥𝑥𝑋𝑋𝑑𝑑 (1) based on Nguen-Widrow method the 𝑥𝑥𝑋𝑋,𝑗𝑗𝑋𝑋 (𝑡𝑡) ∈ (𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 , 𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 ) is jth component of the particle X position; Initialize 𝑦𝑦 𝑑𝑑 (𝑡𝑡) = 𝑥𝑥𝑋𝑋𝑑𝑑 (1), 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 𝑑𝑑 (𝑡𝑡) = 𝑥𝑥𝑋𝑋𝑑𝑑 (1),𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑗𝑗𝑑𝑑 (𝑡𝑡) is jth component of a global best swarm’ position in encoded dimension d. End For. End For. Step 2. For ∀𝑡𝑡 ∈ {1, 𝑇𝑇} do: For ∀𝑋𝑋 ∈ {1, 𝑆𝑆} do: 𝑑𝑑 (𝑡𝑡)

𝑑𝑑 (𝑡𝑡−1)

If 𝑓𝑓(𝑥𝑥𝑋𝑋 𝑋𝑋 (𝑡𝑡)) < 𝑓𝑓 �𝑦𝑦𝑋𝑋 𝑋𝑋

If

𝑑𝑑 (𝑡𝑡)

(𝑡𝑡−1) 𝑑𝑑�

𝑓𝑓 �𝑥𝑥𝑋𝑋 𝑋𝑋 (𝑡𝑡)� > 𝑓𝑓 �𝑦𝑦𝑋𝑋 𝑥𝑥

(𝑡𝑡)

(𝑡𝑡)

(𝑡𝑡 − 1)�then Do: 𝑦𝑦𝑋𝑋𝑑𝑑𝑋𝑋 (𝑡𝑡) = 𝑥𝑥𝑋𝑋𝑑𝑑𝑋𝑋 (𝑡𝑡) (𝑡𝑡 − 1)�

𝑑𝑑 (𝑡𝑡)

then

𝑑𝑑 (𝑡𝑡)

𝑑𝑑𝑋𝑋 (𝑡𝑡) = 𝑑𝑑𝑋𝑋 (𝑡𝑡 − 1)

else𝑑𝑑𝑋𝑋 (𝑡𝑡) = 𝑑𝑑𝑋𝑋 (𝑡𝑡) End If. else 𝑦𝑦𝑋𝑋 𝑋𝑋 (𝑡𝑡) = 𝑦𝑦𝑋𝑋 𝑋𝑋 (𝑡𝑡 − 1) End If. (𝑡𝑡)

𝑑𝑑 (𝑡𝑡) 𝑑𝑑 (𝑡𝑡) 𝑑𝑑 If 𝑓𝑓(𝑥𝑥𝑋𝑋 𝑋𝑋 (𝑡𝑡)) < 𝑚𝑚𝑚𝑚𝑚𝑚(𝑓𝑓(𝑦𝑦𝑋𝑋 𝑋𝑋 (𝑡𝑡 − 1)) ,  𝑚𝑚𝑚𝑚𝑚𝑚 (𝑓𝑓(𝑥𝑥𝑝𝑝 𝑋𝑋 (𝑡𝑡))))  1≤𝑝𝑝 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼 OR𝑓𝑓 �𝑥𝑥𝑔𝑔�𝑑𝑑 � < 𝜀𝜀𝑐𝑐 ) then Stop. End If. 𝑋𝑋 (𝑡𝑡)�

Step

3.

𝑑𝑑 (𝑡𝑡) �.While 𝑋𝑋 (𝑡𝑡)� 𝑡𝑡 𝑋𝑋,𝑡𝑡

𝑋𝑋 𝐸𝐸 = 𝑓𝑓 �𝑥𝑥𝑔𝑔�𝑑𝑑

(𝐼𝐼 < 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼

OR𝐸𝐸 > 𝜀𝜀𝑐𝑐 OR

𝑚𝑚𝑚𝑚𝑚𝑚𝑠𝑠̃ (𝑇𝑇 𝑡𝑡 − 𝑃𝑃𝑋𝑋,𝑡𝑡 )⁄𝑚𝑚𝑚𝑚𝑚𝑚𝑠𝑠̆ (𝑇𝑇 − 𝑃𝑃 ) ≈ 1) 𝛥𝛥𝑊𝑊𝐼𝐼 = [𝐽𝐽𝐼𝐼𝑇𝑇 𝐽𝐽𝐼𝐼 + 𝜇𝜇 ∗ 𝑋𝑋]−1 𝐽𝐽𝐼𝐼𝑇𝑇 𝐸𝐸, 𝐽𝐽𝐼𝐼 the Jacobian matrix, 𝜇𝜇 is the learning rate.

is

Advances in the Artificial Intelligence Models for Photovoltaic Systems 𝑑𝑑 (𝑡𝑡)

11

𝑑𝑑 (𝑡𝑡)

𝑋𝑋 𝑋𝑋 Step 4. Calculate 𝑊𝑊𝐼𝐼 = 𝑊𝑊𝐼𝐼 + 𝛥𝛥𝑊𝑊𝐼𝐼 , 𝑥𝑥 ′𝑔𝑔(𝑑𝑑 , 𝐸𝐸 ′ = 𝑓𝑓(𝑥𝑥 ′𝑔𝑔(𝑑𝑑 ) 𝑋𝑋 (𝑡𝑡)) 𝑋𝑋 (𝑡𝑡)) If 𝐸𝐸 ′ < 𝐸𝐸 then 𝑊𝑊𝐼𝐼 = 𝑊𝑊𝐼𝐼 + 𝛥𝛥𝑊𝑊𝐼𝐼 ;𝜇𝜇 = 𝜇𝜇𝜇𝜇; 𝐸𝐸 ′ = 𝐸𝐸; Go to step 3 else 𝜇𝜇 = 𝜇𝜇/𝛽𝛽; go to step 4. Step 5. For ∀𝑋𝑋 ∈ {1, 𝑆𝑆} do: For ∀𝑗𝑗 ∈ {1, 𝑑𝑑𝑋𝑋 (𝑡𝑡)} do: Generate u and k as U(0,1); 𝛾𝛾 and G has Laplace distribution; 𝑖𝑖𝑖𝑖 𝑘𝑘 ⩾ 0.5 𝑡𝑡ℎ𝑒𝑒𝑒𝑒 𝛼𝛼 = 2 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝛼𝛼 = 1; 𝛽𝛽 is contraction–expansion coefficient; 𝑑𝑑 (𝑡𝑡) 𝐺𝐺 ⋅ 𝑦𝑦𝑋𝑋,𝑖𝑖𝑋𝑋 (𝑡𝑡) + 𝛾𝛾 ⋅ 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑗𝑗𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 (𝑡𝑡) 1 𝑑𝑑𝑋𝑋 (𝑡𝑡) 𝑥𝑥𝑋𝑋,𝑗𝑗 (𝑡𝑡 + 1) = + (−1)𝛼𝛼 𝛽𝛽 ⋅𝑙𝑙𝑙𝑙 𝑙𝑙𝑙𝑙 � � ⋅ 𝐺𝐺 + 𝛾𝛾 𝑢𝑢 𝑑𝑑𝑋𝑋 (𝑡𝑡) 𝑆𝑆 ∑𝑖𝑖=1 𝑦𝑦𝑔𝑔(𝑑𝑑),𝑖𝑖 (𝑡𝑡)  𝑑𝑑 (𝑡𝑡) − 𝑥𝑥𝑋𝑋,𝑗𝑗𝑋𝑋 (𝑡𝑡) ∣, ∣ 𝑆𝑆 𝑑𝑑 (𝑡𝑡) 𝑑𝑑 (𝑡𝑡) 𝑥𝑥 𝑋𝑋 (𝑡𝑡 + 1) if 𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 ≤ 𝑥𝑥𝑋𝑋,𝑗𝑗𝑋𝑋 (𝑡𝑡 + 1) ≤ 𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 𝑑𝑑 (𝑡𝑡) 𝑥𝑥𝑋𝑋,𝑗𝑗𝑋𝑋 (𝑡𝑡 + 1) ← � 𝑋𝑋,𝑗𝑗 � 𝑈𝑈(𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 , 𝑋𝑋𝑚𝑚𝑚𝑚𝑚𝑚 )else In other encoded dimensions ∀𝑑𝑑 ∈ {𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 } − 𝑑𝑑𝑋𝑋 (𝑡𝑡)do updates 𝑑𝑑 𝑑𝑑 𝑥𝑥𝑋𝑋,𝑗𝑗 (𝑡𝑡 + 1) = 𝑥𝑥𝑋𝑋,𝑗𝑗 (𝑡𝑡). End For.

Compute𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡 + 1) = 1

𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡)� ⋅𝑙𝑙𝑙𝑙 𝑙𝑙𝑙𝑙 � �, 𝑢𝑢

𝐺𝐺⋅𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡)+𝛾𝛾⋅𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝐺𝐺+𝛾𝛾

∑𝑆𝑆𝑖𝑖=1 𝑑𝑑~ 𝑔𝑔(𝑑𝑑),𝑖𝑖 (𝑡𝑡) 

+ (−1)𝛼𝛼 𝛽𝛽 ⋅ �

𝑆𝑆

−

𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡 + 1) if 𝐷𝐷𝑚𝑚𝑚𝑚𝑚𝑚,𝑞𝑞 ≤ 𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡 + 1) ≤ 𝐷𝐷𝑚𝑚𝑚𝑚𝑚𝑚,𝑞𝑞 𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡 + 1) = � �, End For. 𝑑𝑑𝑋𝑋,𝑞𝑞 (𝑡𝑡)else End For. The modified MD QPSO elaborates an inter-dimensional quantumbehaved swarm’s search for both positional and dimensional optimum and creates a global best swarm’ encoded dimension 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑. The modified MD QPSO automatically generates an optimum architecture of the MFFN – 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔.

The intelligent times series forecasting framework Given a set of r-dimensional inputs I and desired outputs T, (we assume that the T is one-dimension vector), we split the data s=(I,T) into two equal size sets, by computing the mean over s and sorting the data based on the distance from this mean. Every other data point belonged to a training set 𝑠𝑠̃ , the rest for a test set 𝑠𝑠̆ . This splitting procedure is the same one from GMDH algorithm (Engel, 2020). We calculated the estimate of Lipschitz constant of the data sets as follows:

12

Ekaterina A. Engel 𝛬𝛬𝐷𝐷 = 𝑚𝑚𝑚𝑚𝑚𝑚 𝑖𝑖≠𝑗𝑗

‖𝑇𝑇 𝑖𝑖 −𝑇𝑇 𝑗𝑗 ‖ ‖𝐼𝐼 𝑖𝑖 −𝐼𝐼 𝑗𝑗 ‖

.

(1)

The MFNN includes: a two-layer RNN, fuzzy rules and k two-layered RNNs. The MFNN architecture’s parameters (a hidden neurons’ number– ������ 𝑑𝑑ℎ𝑗𝑗 ∈ 1. . 𝐻𝐻, corresponded biases and weights, a number of RNN’s delays – ������ 𝑑𝑑𝑞𝑞 ∈ ������ 1. . 𝐿𝐿, 𝑗𝑗 ∈ 1. . 𝐿𝐿) were encoded into particles X. 𝑗𝑗

We calculated the dimension of particle X as follows:

𝐷𝐷�𝑑𝑑ℎ,𝑞𝑞 � = ∑k+2 𝑗𝑗=1 𝑑𝑑ℎ𝑗𝑗 ∙ �𝑟𝑟 + 3 ∗ 𝑑𝑑𝑞𝑞𝑗𝑗 �.

(2)

We encoded the dimension component of particle X as

𝑑𝑑 = �𝑑𝑑ℎ1 , 𝑑𝑑ℎ2 , 𝑑𝑑ℎ3 , 𝑑𝑑𝑞𝑞1 , 𝑑𝑑𝑞𝑞2 , 𝑑𝑑𝑞𝑞3 � ∈ {𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = (1,1,1,1,1,1), 𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = (𝐻𝐻, 𝐻𝐻, 𝐻𝐻, 𝐿𝐿, 𝐿𝐿, 𝐿𝐿)}. (3)

We calculated the estimate of Lipschitz constant of an RNN as follows:

𝛬𝛬ℎ ≤ 𝐻𝐻√𝑟𝑟 + 2 ∗ 𝐿𝐿. (4)

Then we evaluated the maximum number of hidden units 𝐻𝐻as follows:

𝐻𝐻 ≥ 𝛬𝛬𝐷𝐷 ⁄√𝑟𝑟 + 2 ∗ 𝐿𝐿. (5)

We defined a fitness function based on the Chebyshev criterion as follows: 𝑓𝑓(𝑋𝑋) = 𝑚𝑚𝑚𝑚𝑚𝑚𝑠𝑠̃ (𝑇𝑇 𝑡𝑡 − 𝑃𝑃 𝑋𝑋,𝑡𝑡 )/𝑇𝑇 𝑡𝑡 , (6)

where 𝑃𝑃𝑋𝑋,𝑡𝑡 is the value which generated based on MFNN with architecture X. We implemented an MFNN and its life cycle, which includes automatic creation and self-adaptation as an intelligent framework based on the authors’ software (The module of the modified fuzzy neural net. M). This intelligent framework provides the automatic creation of the optimum architecture of an MFNN with regard to a task’s complexity. The intelligent times series forecasting framework includes a splitting procedure from GMDH algorithm, evaluation of the maximum number of hidden units 𝐻𝐻 by eq. (5) and the modified MD QPSO.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

13

ML Frameworks All the above-mentioned ML methods and algorithms were implemented as an ML Framework, which represents a tool to create a smart PV system (Nguyen et al., 2019). Table 2 shows the comparison of popular ML frameworks. Google, Facebook, and Microsoft developed most of the DL frameworks that support ONNX, namely PyTorch, TensorFlow, Caffe2, Microsoft CNTK, and MXNet. The high-level DL wrapper libraries such as Keras, Tensor Layer, and Gluon are developed on top of the DL frameworks. They provide a simpler but more computationally expensive way for smart PV system creation. Chainer, Theano, Deeplearning4, and H2O are also appropriate DL libraries and frameworks for smart PV system creation. Table 2. Comparison of ML frameworks Name TensorFlow Microsoft CNTK Caffe2

PyTorch

Keras

Advantages Open-source, API-oriented, crossplatform, ML/DM toolbox implements many ML methods. Open-source, fast-evolving, supports ONNX, supported by Microsoft. Cross-platform, supports mobile platforms, supports ONNX. Dynamic computational graph. Automatic implementation of ML models. Supports ONNX. Open-source provides backend tools from Google and Microsoft. Detailed specification. API for DL. Quick implementation of DL models (e.g., TensorFlow, Theano, CNTK).

Disadvantages The code is not flexible. Lack of documentation. Toolbox oriented for academic usage. Limited facilities for mobile platforms. Complex as compared to PyTorch. Without dynamic graph computation. Absence of monitoring and visualization tools like a tensor board Modularity and simplicity make gains at the expense of flexibility. Limited facilities to create a new architecture.

The ML frameworks provide an automatic MC phase of an ML model, including validating (Figure 1). An ML PV system can be implemented as software based on an ML framework that supports ONNX. Such implementation will provide flexibility and all an ML framework’s advantages for a developed ML PV system.

14

Ekaterina A. Engel

Open Resources for AI Research in a PV System The open solar energy data sources, including big data, provide the development of cutting-edge AI models in solar energy. The GitHub repositories (Avila et al., 2022) are implementations of maximum power point tracking (MPPT) systems (Avila et al., 2022) and management of cities’ demand/load (https://github.com/intelligentenvironments-lab/CityLearn) based on an open-source Gym toolkit (https://gym.openai.com). An open-source tool pymgrid (https://github.com/Total-RD/pymgrid) provides the creation and simulation of various microgrids. Octave (https://octave.org/download) and Scilab (https://www.scilab.org/download/scilab-6.1.1) are open sources that are compatible with MATLAB. Table 3 presents a brief description of the open datasets to implement and validate AIPV systems Table 3.The open datasets to implement and validate AIPV systems Open Dataset Duke California Solar Array Dataset (Bradbury et al., 2016) SOLETE (The_SOLETE_dataset/1704 0767) Desert Knowledge Australia Center Dataset (DKASC, Alice spring)

Girasol (Terrén-Serrano et al., 2021)

Data source Location

Description Over 400 km2 of imagery and 16,000 hand-labeled solar arrays

City: Roskilde, Denmark. Latitude and longitude: 55.6867, 12.0985 -

Albuquerque, USA

ESOLMET-IER Dataset (ESOLMET-IER instituto de energíasrenovables) The National Solar Radiation Data Base (NSRDB) (Sengupta et al., 2018)

Institute of Renewable Energies UNAM, station “ESOLMET-IER”

Photovoltaic Thermal Images Dataset (Pierdicca et al., 2022)

66 MW PV Tomboruk

Pecan Street Dataset (Miranda et al., 2021)

-

The USA and neighboring countries plant

in

Meteorological and active power 15 months dataset from PV array Data of PV systems spanning multiple types, ages, models, and configurations A meteorological (10 min sampling interval), insolation (a sampling rate ranging from 4 to 6 samples per second), and images (sampling interval of the cameras is 15 s) 242 days (of 3 years) dataset Solar metric and meteorological dataset Solar insolation and meteorological 23 years dataset Thermal images of PV arrays with the presence of one or more anomaly cells and their respective masks 1300 customer loads one-year dataset

Advances in the Artificial Intelligence Models for Photovoltaic Systems

15

Machine Learning Applications for a PV System This section presents a review of research studies that have been published mostly in the last five years on the topic of AI applications for a PV system. The literature review process elaborates on the articles’ search queries in Scopus/ScienceDirect, IEEE Xplore, ResearchGate, and Google Scholar with the following keywords: machine learning, neural networks, DL, PV, and PV system. We focused on four important tasks’ categories in the PV systems, as shown in Figure 3: design, forecasting, maintenance, and control. Figure 3 identifies the number of publications devoted to AI for a PV system’s design, forecasting, maintenance, and control that have been published mostly during the last five years.

Figure 3. AI models for a PV system’s design, forecasting, maintenance, and control.

16

Ekaterina A. Engel

We prepared the data based on the considerable contributions from the most cited journals. We have not covered cyber security in a PV system since it was covered in-depth in study (Kurukuru et al., 2021). Figure 4 reflects the number of publications devoted to CNN, ANN, and RNN models for PV systems that have been published mostly in the last five years. Figure 4 also presents the various types of feature spaces to create a smart PV system based on an ML method. It specifies the essential preprocessing and ML models to create a smart PV system (Ha et al., 2020). We proposed a simple but effective pipeline scheme of an implementation (implementation step in Figure 1) of a ML PV system. Figure 5 shows this simple scheme of a PV system based on ML methods for a PV system’s design, forecasting, maintenance, and control. The SCADA system is able to integrate a PV system and ML models into an ML PV system based on software that implements ML PV models and integrates with SCADA through API. These ML models for a PV array’s design, forecasting, maintenance, and control are implementations of a basic ML model class, which is represented in Figure 5 as a UML class diagram. A method “Train” of a basic ML model class implements the MC phase, including validating. Thus, the impact of the ML models based on the proposed scheme (Figure 5) on a PV system value chain will mostly be associated with the cost of software development and maintenance. Most AIPV models can be implemented on a solar plant based on the proposed scheme. Such implementation will provide flexibility and all ML framework’s advantages for the developed AI PV system and its digital transformation into a smart PV system, which we outlined in last Section.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

Figure 4. Classification of ML models for a PV system.

Figure 5. PV system based on ML methods.

17

18

Ekaterina A. Engel

AI Models for Design of the PV systems The optimal design of a PV system is a very complex task that requires the fulfillment of models for a PV system’s components as well as the usage of global optimizers. Parameter Identification in a PV System The parameter extraction models for the single (SDM), double (DDM), or triple diode solar cell model (TDM) with RMSE as the performance metric are highly demanded for simulation and fault detection of a PV system. There are many heuristic search algorithms, including bioinspired, that were adapted to solve the parameter identification task of the different solar cell models. Table 4 displays a brief comparison of the AI parameter identification models from studies. Table 4. Comparison of the parameter identification models Algorithm ANN (Abdellatif et al., 2022) ANN (Awadallah, 2014) Flexible PSO (Ebrahimi et al., 2019) Whale optimization (Elazab et al., 2017)

Tree-growth-based optimization (TG)

(Diab et al., 2020)

Memetic adaptive differential evolution (MD) (Li et al., 2019) Artificial Bee Colony (ABC) (Xu et al., 2021)

Outperforms

Diode Model

RMSE

RBF-NN

SDM

Low

ANFIS

SDM

Low

Classical PSO

SDM DDM

Classical PSO

DDM

Two-step Linear Least-Squares (TSLLS) method, Reduced forms RF, Artificial bee swarm optimization (ABSO), Harmony search-based algorithm (HS), Particle swarm optimization (PSO) algorithm, Genetic algorithm (GA), analytical 5-point method (An.5-Pt), the Lambert W (LW) function, Newton method, Conductance method, and pattern search

Modera te

SDM

High

GA

SDM

Low

Classical ABC

SDM DDM

and

and

High

Low

Advances in the Artificial Intelligence Models for Photovoltaic Systems Algorithm

Outperforms

JAYA-based (Yu et al., 2022; Yu et al., 2019)

Chaos Game Optimization (CGO) (Zellagui et al., 2022)

Supply-DemandBased (Shaheen et al., 2022)

Covariance matrix adaptation evolution strategy (CMAES), Grey Wolf Optimizer (GWO), Teaching-learning-based artificial bee colony (TLABC), Transactional agents for pervasive computing (TAPSO), MLbased stealing attack methodology (MLBSA), Generalized oppositional teaching learning-based optimization (GOTLBO) W, TG, MD, applied chaotic reproduction optimization (CARO) (Xu et al., 2021), modified simplified swarm optimization algorithm (MSSO) (Lin et al., 2017), Cuckoo search algorithm (CSA) (lang et al., 2018), Biogeography optimization algorithm-based heterogeneous cuckoo search (BBO-HCS) algorithm (Chen and Yu, 2019) Backtracking Search Algorithm, Grey Wolf Optimizer, Bernstein–Levy Search Differential Evolution Algorithm, Crow Search Optimizer, and Manta Ray Foraging Optimizer

19

Diode Model

RMSE

SDM

Low

SDM

Low

TDM

Low

In the work by the researchers (Pillai et al., 2018), the parameter identification models for 17 different industrial solar cells/modules are reported. The hybrid bee pollinator flower pollination algorithm (BPFPA) (Ram et al., 2017) has the lowest RMSE and highest convergence as compared to all 21 reviewed parameter identification metaheuristic algorithms. Table 5 summarizes the comparative results of different to set benchmarks for the performance comparison of the parameter identification models based on different metaheuristic algorithms for the 57 mm dia RTC France solar cell. Table 5. Performance comparison of the parameter identification models for the 57 mm dia RTC France solar cell (Pillai et al., 2018) Single Diode Model RMSE Algorithm BPFPA (Pillai et al., 7.27 × 10−4 2018) 7.84 × 10−4 9.45 × 10−4 9.86 × 10−4

FPA (Pillai et al., 2018) MPCOA (Pillai et al., 2018) STLBO (Pillai et al., 2018)

Sl. No. 1 2 3 4

Double Diode Model Algorithm BPFPA (Pillai et al., 2018) FPA (Pillai et al., 2018) MPCOA (Pillai et al., 2018) STLBO (Pillai et al., 2018)

RMSE 7.23 × 10−4 7.73 × 10−4 9.22 × 10−4 9.82 × 10−4

20

Ekaterina A. Engel

Table 5. (Continued) Single Diode Model RMSE Algorithm R-JADE (Pillai et al., 9.86 × 10−4 2018) TVIWAC PSO (Pillai et 9.86 × 10−4 al., 2018) BMO (Pillai et al., 9.86 × 10−4 2018) ABC+NMS (Pillai et 9.86 × 10−4 al., 2018) 9.86 × 10−4

ABC (Pillai et al., 2018)

9.86 × 10−4 9.86 × 10−4 9.8602 10−4 9.8602 10−4 9.8602 10−4 9.8602 10−4 9.8602 10−4 9.8602 10−4 9.860219 10−4 9.86022 10−4 9.86023 10−4 9.8605 10−4 9.8607 10−4 9.8625 10−4 9.8665 10−4

× × × × × × × × ×

BBO-M (Pillai et al., 2018) LM + SA (Pillai et al., 2018) TLABC (Yu et al., 2022) TAPSO (Yu et al., 2022) MLBSA (Yu et al., 2022) GOTLBO (Yu et al., 2022) PGJAYA (Yu et al., 2022) HAJAYADE (Yu et al., 2022) CGO (Zellagui et al., 2022) BBO-HC (Chen and Yu, 2019)

Sl. No. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

CSA (Kang et al., 2018)

20

CMM-DE/BBO (Yu at al., 2019)

21

MSSO (Lin et al., 2017)

22

IJAYA (Yu et al., 2022)

23

CARO (Xu et al., 2021)

24

9.87 × 10−4

PSA (Pillai et al., 2018)

25

9.89 × 10−4

IADE (Pillai et al., 2018)

26

× × × ×

Double Diode Model Algorithm R-JADE (Pillai et al., 2018) ABC+ NMS (Pillai et al., 2018) TAPSO (Yu et al., 2022) MLBSA (Yu et al., 2022) PGJAYA (Yu et al., 2022) GOTLBO (Yu et al., 2022) BMO (Pillai et al., 2018) BB0-M (Pillai et al., 2018) ABSO (Pillai et al., 2018) TLABC (Yu et al., 2022) ABC (Pillai et al., 2018) IGHS (Pillai et al., 2018) IJAYA (Yu et al., 2022) JAYA (Yu at al., 2019) CMAES (Yu et al., 2022) CLPSO (Yu at al., 2019) CMM-DE/BBO (Pillai et al., 2018) DE/BBO (Yu at al., 2019) BLPSO (Yu at al., 2019) GGHS (Pillai et al., 2018) GWO (Yu et al., 2022) HS (Pillai et al., 2018)

RMSE 9.82 × 10−4 9.82 × 10−4 9.8269 × 10−4 9.8285 × 10−4 9.8298 × 10−4 9.8299 × 10−4 9.83 × 10−4 9.83 × 10−4 9.83 × 10−4 9.8407 × 10−4 9.86 × 10−4 9.86 × 10−4 9.8631 × 10−4 9.8934 × 10−4 9.9015 × 10−4 9.9894 × 10−4 1.0088 × 10−3 1.0255 × 10−3 1.0628 × 10−3 1.07 × 10−3 1.1429 × 10−3 1.26 × 10−3

Advances in the Artificial Intelligence Models for Photovoltaic Systems Single Diode Model RMSE Algorithm 9.8946 × JAYA (Yu at al., 2019) 10−4 GGHS (Pillai et al., 9.91 × 10−4 2018) ABSO (Pillai et al., 9.91 × 10−4 2018) IGHS (Pillai et al., 9.93 × 10−4 2018) 9.95 × 10−4 HS (Pillai et al., 2018) 9.9633 × CLPSO (Yu at al., 10−4 2019) 9.9922 × DE/BBO (Yu at al., 10−4 2019) 1.0023 × GWO (Yu et al., 2022) 10−3 1.0272 × BLPSO (Yu at al., 10−3 2019) 1.70 × 10−3 SA (Pillai et al., 2018)

Sl. No.

Double Diode Model Algorithm

RMSE

27

SA (Pillai et al., 2018)

N. S

28

PSO (Pillai et al., 2018)

N. S

21

29 30 31 32 33 34 35 36

Summarizing, we highlight a need to assess more benchmarks for a performance comparison of the parameter identification models including ML methods.

Sizing of a PV system Within the research literature, a whole array of differing sizing methods for a PV system has been proposed. These sizing methods of a PV system are classified as intuitive, numerical, and analytical algorithms. The intuitive algorithms do not provide effectiveness and reliability. The numerical algorithms require a long time series of insolation. Many of the analytical algorithms use a concept of the system’s reliability or loss of load probability. AI models provide an estimation of the optimal number of panels, storage capacity of batteries, tilt, and azimuth angles for a PV system. Moreover, several ML models have been developed to size a PV system. Table 6 shows a brief comparison of ML sizing methods of a PV system. Summarizing, we highlight a need to assess more benchmarks for a performance comparison of the PV sizing ML models. In addition, DL methods, including RNN, that extract knowledge from time series and effectively approximate insolation and load under small disturbances of a PV system dynamic, including degradation, are promising alternatives.

22

Ekaterina A. Engel

Table 6. Comparison of ML sizing methods Sizing Method Generalized RNN (Khatib et al., 2014)

Dataset Meteorological and load demand dataset from five Malaysian sites

Performance

Contribution

MAE% is 0.6%

CNN creates semantic segmentation SolarMapper (https://solarmapp er.anl.gov/)

CNN (Malof et al., 2019)

Duke California Solar Array dataset

Object-based performance metric is 0.76

DNN framework (Mason et al., 2020)

Behind-the-meter load dataset that includes erroneous and mislabeled training data

MFNN (Engel, 2020)

Two-year dataset of total insolation, meteorological parameters which was collected at the site of Abakan

MAE% in estimation of a PV tilt and azimuth values are 10.1% and 2.8%, correspondingly

ML optimization method based on ANN and heuristic optimizers (Zhou et al., 2020)

One-month datasets of meteorological parameters which were collected at the different climatic China regions

-

MAE% is 0.6% which is superior to PSO

Automatic creation, self-adaptation MFNN based on the authors’ software

The annual equivalent overall output energy increased by 4.48% as compared to a Taguchi standard orthogonal array

Within the application of smart cities researchers design a renewable system that includes solar-toelectricity conversion.

AI Models for Insolation and Power forecasting of PV systems Energy production of a PV system is highly dependent on weather conditions such as insolation and temperature. Thus, it is difficult to balance the production and consumption of the electric grid with integrated PV systems where production levels fluctuate. In case of a deviation from an hourly plan schedule of solar plant power, the energy market charges penalties. Hence, many AI methods have been implemented to forecast insolation and the output power from a PV system. Figure 6 presents specifics of the energy market to forecasting and classification of AI forecasting models based on a forecasting horizon (Kurukuru et al., 2021; Alkhayat and Mehmood, 2021). The surveys of insolation and power forecasting of a PV system in (Nespoli et al., 2019; Khan et al., 2020; Grimaccia et al., 2017; Omar et al.,

Advances in the Artificial Intelligence Models for Photovoltaic Systems

23

2016; Das et al., 2018; Akhter et al., 2019; Notton et al., 2019) appraise various approaches and methods to increase the performance of forecasting models under uncertainties. According to the reviews, ANNs are the most popular method for forecasting, as they are easy to implement and quite effective as compared to classical methods, such as conventional autoregressive integrated moving average (ARIMA), etc.

Figure 6. Specifics of the energy market to forecasting and classification of AI forecasting models.

The intelligent two days ahead hourly PV array power forecasting Model We generated the intelligent two days ahead hourly PV array power forecasting model based on the developed intelligent times series forecasting framework (Engel and Engel, 2020). We fulfilled the intelligent two days ahead hourly PV array power forecasting model based on the MFNN which includes: an RNN, fuzzy rules and RNNs. We calculate the total rate of insolation GC striking a PV array on a clear day as follows: 𝐺𝐺𝐺𝐺 = 𝐴𝐴𝑒𝑒 −𝑘𝑘𝑘𝑘 (𝑐𝑐𝑐𝑐𝑐𝑐 𝛽𝛽 𝑐𝑐𝑐𝑐𝑐𝑐( 𝜙𝜙𝑠𝑠 − 𝜙𝜙𝐶𝐶 ) 𝑠𝑠𝑠𝑠𝑠𝑠 𝛴𝛴 + 𝑠𝑠𝑠𝑠𝑠𝑠 𝛽𝛽 𝑐𝑐𝑐𝑐𝑐𝑐 𝛴𝛴 + (𝐶𝐶 + 𝑐𝑐𝑐𝑐𝑐𝑐 𝛴𝛴)/2 + 𝑝𝑝(𝑠𝑠𝑠𝑠𝑠𝑠 𝛽𝛽 + 𝐶𝐶)(1 − 𝑐𝑐𝑐𝑐𝑐𝑐 𝛴𝛴)/2) (7)

Where m – the air mass, β – the altitude angle, φS – the solar azimuth angle, φC – the PV module azimuth angle, p – the reflection factor, 𝛴𝛴 – the PV module tilt angle, C – the sky diffuse factor, A and k – parameters related to the Julian day number. The surface insolation fluctuates accordingly with the cloudiness’ dynamic (Figure 7).

24

Ekaterina A. Engel

1200

1000

800

Irradiance, W/m2

Extraterrestrial irradiance Surface irradiance 600

400

200

0 5AM

05/16/16

11PM 5AM

05/17/16

11PM 5AM

05/18/16

11PM 5AM

05/19/16

11PM 5AM

05/20/16

11PM

Figure 7. The extraterrestrial insolation and the surface insolation at the site of Abakan.

We define a clear-sky index as follows: 𝐶𝐶 = 𝐺𝐺𝐺𝐺/𝐺𝐺𝐺𝐺, where Gs is the surface insolation, Gc is the clear-sky model’s insolation. Figure 8 shows that the clear-sky index C is big and has similar shape on sunny days (05/18/16, 05/19/16) at the site of Abakan. In contrast, C is smaller and has more fluctuations on cloudy days (05/16/16, 05/17/16). 1

Clear-sky index

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 5AM

05/16/16

11PM 5AM

05/17/16

11PM 5AM

05/18/16

11PM 5AM

05/19/16

11PM 5AM

05/20/16

11PM

Figure 8. The clear-sky index at the site of Abakan.

We fulfilled the intelligent two days ahead hourly PV array power forecasting model based on the data sht= (Iht= (G0ht, C h t-2, P h t-2, lht, aht, Rht, Wht, dht), Tht =Pht), (8)

Advances in the Artificial Intelligence Models for Photovoltaic Systems

25

where G0ht is the extraterrestrial insolation, Phtis the power from a PV array, Ph t-2is the historical data of the power from a PV array, Cht-2 is the historical data of clear-sky index, lht is the cloudiness (%),Rht is the pressure, Whtanddht are the wind speed and the wind direction, respectively, aht is the ambient temperature, h=10. .18, t=1. .1309. It is to emphasize that lht,Rht, Wht, dht, aht are daily average parameters of the weather forecast. The data (8) was collected at the Abakan from 03/16 through 09/19. We generated the intelligent two days ahead hourly PV array power forecasting model based on the developed intelligent times series forecasting framework. We described the fulfillment of the intelligent two days ahead hourly PV array power forecasting model briefly as follows. Step 1. All samples of the data (7) 𝑠𝑠 𝑖𝑖 were classified into two groups according to hour’s state of cloudiness: 𝐴𝐴1is sunny hour (𝐶𝐶 𝑖𝑖1 = 1), 𝐴𝐴2 is cloudy hour (𝐶𝐶 𝑖𝑖2 = −1). This classification generates vector with elements 𝐶𝐶 𝑖𝑖 . Step 2. We created the two-layer network: 𝑌𝑌�𝑠𝑠 𝑖𝑖 � . The vector 𝑠𝑠 𝑖𝑖 was the network's input. The vector 𝐶𝐶 𝑖𝑖 was the network's target. We formed membership function 𝜇𝜇𝑗𝑗 (𝑠𝑠)based on the two-layer network 𝑌𝑌�𝑠𝑠 𝑖𝑖 � asfollows 𝜇𝜇1 (𝑠𝑠 𝑖𝑖 ) = �

𝑌𝑌(𝑠𝑠 𝑖𝑖 ), 𝑖𝑖𝑖𝑖 𝑌𝑌(𝑠𝑠 𝑖𝑖 ) ≥ 0 , 0, 𝑖𝑖𝑖𝑖 𝑌𝑌(𝑠𝑠 𝑖𝑖 ) < 0

𝜇𝜇2 (𝑠𝑠 𝑖𝑖 ) = �

�𝑌𝑌(𝑠𝑠 𝑖𝑖 )�, 𝑖𝑖𝑖𝑖 𝑌𝑌(𝑠𝑠 𝑖𝑖 ) < 0 0, 𝑖𝑖𝑖𝑖 𝑌𝑌(𝑠𝑠 𝑖𝑖 ) ≥ 0

(9)

This step provides the fuzzy sets 𝐴𝐴𝑗𝑗 , (𝐴𝐴1 is sunny hour, 𝐴𝐴2 is cloudy hour) 1. .2. with membership function 𝜇𝜇𝑗𝑗 (𝑠𝑠), 𝑗𝑗 = ����� Step 3. We created the MFNN based on the data (7). The MFNN includes 𝑖𝑖 𝑌𝑌�𝑠𝑠 � and two RNNs 𝐹𝐹𝑗𝑗 . The MFNN architecture’s parameters (number of nodes in hidden layer, corresponded weights and biases) have been encoded into particles X. We calculate the maximum number of hidden units 𝐻𝐻 by eq. (5). We calculate the dimension component of particle X by eq. (2), r=8. To make the intelligent two days ahead hourly power forecasting from the PV array model become adaptive, it needs to have some idea of how the actual hourly power differs from it expected the two days ahead hourly power, so that the RNN 𝐹𝐹𝑗𝑗 can recalibrate its value intelligently during run time, and try to eliminate the constant tracking error. We give the RNN 𝐹𝐹𝑗𝑗 (𝜇𝜇𝑗𝑗 (𝑠𝑠), 𝑠𝑠) an extra input 𝜇𝜇𝑗𝑗 (𝑠𝑠) which corresponds to the value of membership function 𝜇𝜇𝑗𝑗 (𝑠𝑠). This input signal of the RNNs 𝐹𝐹𝑗𝑗 (𝜇𝜇𝑗𝑗 (𝑠𝑠), 𝑠𝑠) will give useful feedback for providing the two days ahead hourly power forecasting value during the

26

Ekaterina A. Engel

dynamically changing times series (8). This forecasting approach does provide a more intelligent algorithm of generating the two days ahead hourly power forecasting value 𝑢𝑢 on the basis of a MFNN. We used modified MD QPSO as an optimization algorithm. We used function (6) as a fitness function for the modified MD QPSO. This step provides trained MFNN 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏(𝑑𝑑ℎ ) which create the forecasted power of the PV array 𝑢𝑢(𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏(𝑑𝑑ℎ )) – best solution 𝑋𝑋 generated by the modified MD QPSO). If-then rules are defined as: 1. .2. П𝑗𝑗 : 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐴𝐴𝑗𝑗 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 = 𝐹𝐹𝑗𝑗 (𝜇𝜇𝑗𝑗 (𝑠𝑠), 𝑠𝑠), 𝑗𝑗 = �����

(10)

Simulation of the trained MFNN briefly can be described as follows. Step 1. Aggregation antecedents of the rules (10) maps input data 𝑠𝑠 into their membership functions and matches data with conditions of rules. These mappings are then activating the 𝑘𝑘 rule, which indicates the 𝑘𝑘 hour’s state of 1. .2. cloudiness and correspondent 𝑘𝑘 RNN 𝐹𝐹𝑘𝑘 (𝜇𝜇𝑗𝑗 (𝑠𝑠 𝑡𝑡 ), 𝑠𝑠 𝑡𝑡 ),  𝑘𝑘 ∈ ����� Step 2. According the 𝑘𝑘 mode the correspondent 𝑘𝑘 RNN 𝐹𝐹𝑘𝑘 (𝜇𝜇𝑗𝑗 (𝑠𝑠 𝑡𝑡 ), 𝑠𝑠 𝑡𝑡 ) (trained based on the data st-2 (8)) generates the two days ahead hourly power forecasting value 𝑢𝑢 = 𝐹𝐹𝑗𝑗 (𝜇𝜇𝑗𝑗 (𝑠𝑠 𝑡𝑡 ), 𝑠𝑠 𝑡𝑡 ). We created the intelligent two days ahead hourly PV array power forecasting model within the developed intelligent times series forecasting framework based on the training set of the data (7) from 03/16 through 04/19. We trained the MFNN using modified MD QPSO with Laplace distribution(MD QPSO-LD). We calculate the maximum number of hidden units 𝐻𝐻 by eq. (2), 𝐻𝐻 = 14. Due to obtain statistical results, we perform 120 modified MD QPSO and PSO runs with following parameters: 𝑛𝑛 = 250 (we use 250 particles), 𝑇𝑇 = 450 (the maximum number of iterations is 450). Figure 9 represents that the modified MD QPSO has confidently more the convergence speed than the modified PSO over training set of the data (8). The modified MD QPSO created the optimum architecture of the MFFN 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 with 𝐷𝐷(𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 = 𝑑𝑑ℎ=(5,7,10),𝑞𝑞=(2,2,2) ) = 336. This MFNN includes the

two-layered RNN 𝑌𝑌�𝑠𝑠 𝑖𝑖 �with five hidden neurons (number of delays is two) and then the two two-layered RNNs 𝐹𝐹𝑘𝑘 (𝜇𝜇𝑘𝑘 (𝑠𝑠), 𝑠𝑠) with seven and ten hidden ����� neurons, correspondently (number of delays is two),  𝑘𝑘 = 1. .2. Figure 10 represents the plot of the measured power from the Abakan solar plant on the test set of the data (8) from 05/19 through 09/19 in comparison to forecasted power of the MFNN and the RNN (two-layered with

Advances in the Artificial Intelligence Models for Photovoltaic Systems

27

ten hidden neurons, the number of delays is two), which we generated by the Levenberg-Marquardt algorithm based on training set of the data (8).

Figure 9. The mean convergence curves. 4500

The actual value of P The forecasted value of P based on RNN The forecasted value of P based on MFNN

4000

3500

3000

Power P, kW

2500

2000

1500

1000

500

0 05/19

06/19

Time

07/19

Figure 10. The forecasted power curves

08/19

09/19

28

Ekaterina A. Engel

Table 7 demonstrates that fitness function (6) of the MFNN in sunny hours is quite small as compared with the RNN, which we generated as two-layered based on data (8) with ten hidden neurons, the number of delays is two. The performance of the MFNN is changing in sunny and cloudy hours (Table 7). Table 7. A two days ahead forecasting of the hourly power from the PV array: comparison of results

The fitness function (6) (%)

MFNN 𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 solution Sunny Cloudy 3,74 4,67

RNN solution Sunny Cloudy 9,82 18,73

Nevertheless, the MFNN effectively tracks the complex dynamics of real measured data in cloudy hours. Table 7 indicates that the MFNN outperforms the RNN, especially in the cloudy hours. The performance of the MFNN trained by the proposed algorithm is superior to the RNN trained by the Levenberg-Marquardt algorithm, especially on cloudiness condition. It is shown that the intelligent two days ahead hourly PV array power forecasting model on the basis of a MFNN is robust to various uncertainties. Unlike popular approaches to nonlinear approximation, a MFNN is used to approximate the times series’ law, which makes it suitable over a wide range of nonlinearities. Compared to standard neural networks, including RNN, the intelligent two days ahead hourly PV array power forecasting model on the basis of a MFNN produces good performance. Based on the fuzzy hour’s state of cloudiness, the created MFNN provides an effective two days ahead hourly PV array power forecasting under various uncertainties. Simulation comparison results for two days ahead forecasting of PV array power demonstrates the effectiveness of the MFNN generated by modified MD QPSO as compared with the RNN, which were trained by the LevenbergMarquardt algorithm.

ML Models for Power Forecasting of PV systems The power forecasting of a solar plant provides safety and effectiveness of grid control. There are mainly three ways to power forecast for a PV system: only historical output power recorded is used, forecasted meteorological parameters are used as input, combination of the historical power data with forecasted meteorological parameters is used.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

29

Recent studies present the ML methods, which effectively forecast a PV system’s power. The study (Liu et al., 2017) reveals that the output power with the insolation and the air temperature has a linear and nonlinear correlation, correspondingly. Recently, researchers have been more interested in the ML application to increase the accuracy of the forecasters (Das et al., 2018; Suresh et al., 2020; Wu et al., 2022; Tovar et al., 2020; Zjavka et al., 2022; Pombo et al., 2022; Pombo and Binder, 2022; Ren et al., 2022; Ahn and park, 2021; Wang et al., 2018; Hossain and Mahmood, 2020; Mittal, 2022; Luo et al., 2022; Abdellatif et al., 2022). The simple (in Das et al., 2018), preprocessing generated normalized insolation; in Wang et al., 2018), preprocessing elaborated k-means) and complex data preprocessing algorithms (in Ren et al., 2022, four CNNs with different filters mine simple features from a sequence of time series; a singlekernel CNN mines the meta features from the simple features) provide for the ML model better performance (Table 7). Due to forecast power, (in Pombo et al., 2022; Pombo and Binder, 2022), researchers integrated a PV-performance model into ML methods such asRF, SVR, CNN, LSTM, and hybrid CNN-LSTM. The results indicated that the proposed ML models provide the best performance regardless of the model’s type and forecasting horizon (Table 7). Table 7 shows that indirect, very short-term forecasting ML models (Das et al., 2018, Tovar et al., 2020) provide higher accuracy as compared to direct ones. Table 7 shows that the dataset’s length has a positive correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is 0.34). Table 8 displays that the forecast horizon has a negative correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is −0.31). Table 8. The performances of the power forecasting ML models Predicting Method Stack-ETR (TF) (Abdellatif et al., 2022) Stack-ETR (MC) (Abdellatif et al., 2022) Stack-ETR (PC) (Abdellatif et al., 2022) Stack-GBDT (Zhang and Zhu, 2022)

The Forecasting Horizon

Dataset’s Length

RMSE (Wh/m2)

RMSE %

1 day

4 years

37.37

-

1 day

4 years

13.95

-

1 day

4 years

20.41

-

1 day

4 years

47.7826

-

30

Ekaterina A. Engel

Table 8. (Continued) Predicting Method RNN-LSTM (TF) (Akhter et al., 2022) RNN-LSTM (MC) (Akhter et al., 2022) RNN-LSTM (PC) (Akhter et al., 2022) XGBoost-DNN (Kumari and Toshniwal, 2021) DPNN (Zjavka, 2020) K-means-AE-CNNLSTM (Zhen, 2020) LSTM-RNN (Abdel-Nasser et al. 2019) LSTM (Zhang et al., 2018) ELM (TF) (Hossain et al., 2017) ELM (MC) (Hossain et al., 2017) ELM (PC) (Hossain et al., 2017) ANN’s ensemble (Omar et al., 2016) MLPNN (Akhter et al., 2019) TDNN + clustering (Akhter et al., 2019) MLFFNN based on BP (Akhter et al., 2019) CNN-Simple (Suresh et al., 2020) Multi-headed CNN (Suresh et al., 2020) CNN-LSTM (Suresh et al., 2020)

5D CNN-LSTM (Tovar et al., 2020)

D-PNN (Zjavka and Snášel, 2022) RF (Pombo et al., 2022) Support Vector Regression (SVR) (Pombo et al., 2022) CNN (Pombo et al., 2022) LSTM (Pombo et al., 2022) Hybrid (Pombo et al., 2022) RF (Pombo and Binder, 2022)

The Forecasting Horizon

Dataset’s Length

RMSE (Wh/m2)

RMSE %

1 day

4 years

39.2

-

1 day

4 years

19.78

-

1 day

4 years

26.85

-

1 day

10 years

51.35

-

1 day

2 weeks

52.8

-

1 day

-

45.11

-

1 day

1 year

82.15

-

1 day 1 day 1 day 1 day

1 year 1 year 1 year

139.3 90.41 59.93 54.96

-

1h

-

5

6.25%

1 day

1 year

160.3

-

1 day

1 year

122

-

1 day

1 year

223

-

1 day

6 years

51

-

1 day

6 years

81

-

1 day 10 min 30 min 60 min 90 min 120 min 150 min 180 min

6 years 1 year 1 year 1 year 1 year 1 year 1 year 1 year

51 0.083 0.22 0.45 0.72 1.05 1.44 2.05

-

1 day

60

-

1h

15 months

-

11.83%

1h

15 months

-

13.71%

1h 1h 1h 24 h

15 months 15 months 15 months 15 months

-

15.27% 14.89% 15.72% 7.58%

Advances in the Artificial Intelligence Models for Photovoltaic Systems Predicting Method

SVR (Pombo and Binder, 2022)

CNN (Pombo and Binder, 2022) LSTM 2022)

(Pombo

and

Binder,

Hybrid (Pombo and Binder, 2022) Quad-kernel deep CNN (QKCNN) (Ren et al., 2022) SVR-RBF (Ahn and Park, 2021) Deep RNN (Ahn and Park, 2021) BackPropagation NN (Wang et al., 2018) LSTM NN (Hossain and Mahmood, 2020) RNN (Hossain and Mahmood, 2020) Generalized regression neural network (GRNN) (Hossain and Mahmood, 2020) Extreme learning machine (ELM) (Hossain and Mahmood, 2020) Transfer learning constrained LSTM (TL + C-LSTM) (Luo et al., 2022) MFNN (Engel, 2020) RFR (Abdellatif et al., 2022) XGB (Abdellatif et al., 2022) DTR (Abdellatif et al., 2022) ADA (Abdellatif et al., 2022) ETR (Abdellatif et al., 2022) Stack-RFR (Abdellatif et al., 2022) Stack-ETR (Abdellatif et al., 2022) Stack-ADA (Abdellatif et al., 2022) Stack-XGB (Abdellatif et al., 2022)

31

The Forecasting Horizon 48 h 72 h 24 h 48 h 72 h 24 h 48 h 72 h 24 h 48 h 72 h 24 h 48 h 72 h

Dataset’s Length 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months 15 months

RMSE (Wh/m2) -

RMSE % 7.75% 7.93% 8.06% 8.21% 8.29% 8.69% 8.86% 9.16% 7.56% 8.08% 8.12% 8.06% 8.69% 8.96%

10 min

-

-

4%

1h 1h

-

10 5

-

1 day

100 days

3.66

-

1 day

3 months

7.1

-

1 day

3 months

9.2

-

1 day

3 months

13.1

-

1 day

3 months

24.1

-

1 day

1 year

8.89

-

2 day 1 day 1 day 1 day 1 day 1 day

3 years 4 years 4 years 4 years 4 years 4 years

43.15 38.96 34.11 36.61 35.52 32.05

20.15% -

1 day

4 years

24.9

-

1 day

4 years

23.09

-

1 day

4 years

24.58

-

1 day

4 years

23.97

-

32

Ekaterina A. Engel

AI Models for Insolation Forecasting of the PV systems AI models for insolation forecasting provide great benefits to smart grid integration and solar plant management. AI insolation forecasting is a necessary step for indirect power forecasting that provides higher accuracy as compared to a direct one. In Table 8, we briefly summarize the insolation forecasting ML models from studies. Table 8 shows that the dataset’s length has a positive correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is 0.34). Table 8 displays that the forecast horizon has a negative correlation with forecast performance (an average correlation coefficient of normalized corresponding columns is −0.31). Summarizing, we highlight a need to assess more datasets and benchmarks for the performance comparison of ML models for insolation and PV system power forecasting. The number of data preprocessing algorithms has a negative correlation with a forecast’s performance. The dataset’s length and forecast horizon have positive and negative correlation with a forecast’s performance, correspondingly. A one-year test dataset is enough to create and validate a robust ML model. Indirect power forecasting provides higher accuracy as compared to a direct one. In addition, DL methods including transformers based on an attention mechanism that hierarchically preprocess and mine knowledge from datasets are promising alternatives.

Uluru (Ayers Rock) in Australia

Abakan, RF Ghardaia, Algeria

Bangladesh

Tetouan, Morocco

Narino state

Bogota

Puerto Merizalde

Sipi

Chajal

Barrancominas

Caruru

Site

0.1471 0.1146 0.0941 0.0850 0.0729 0.0642

-

RMSE [W/m2] 9.1715 7.006 9.1002 6.9222 7.0558 6.2072 7.8242 6.3490 7.9230 6.6222 7.7266 6.3453 42 64 13.59 15.8 0.958 1.14 0.891 21.5

-

MBE [W/m2] 0.9309 3.1310 0.0568 3.0637 0.3947 2.6189 0.6185 2.7263 0.5521 2.8704 0.6464 2.6964 34.709 23.883 -

-

0.91 0.98–0.96 -

R2 0.9962 0.9977 0.9961 0.9977 0.9976 0.9981 0.9972 0.9982 0.9971 0.9979 0.9973 0.9981 -

2 years

Dataset’s Length 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 11 years 3 years 3 years 6 years 6 years 6 years 2 years 3 years 2 years 2 years 2 years 2 years 2 years

Table 9. The performances of the insolation forecasting ML models Model RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) RF (ESOLMET-IER, 2022) ANN (ESOLMET-IER, 2022) LSTM (Narvaez et al., 2021) LSTM (Narvaez et al., 2021) SVM (Brahim et al., 2022) ANN (Brahim et al., 2022) RNN (Faisal et al., 2022) LSTM (Faisal et al., 2022) GRU(Faisal et al., 2022) [88] MFNN (Engel, 2020) LSTM (Guermoui et al., 2022) ShuffleNet (Acikgoz, 2022) SqueezeNet (Acikgoz, 2022) ResNet-18 (Acikgoz, 2022) GoogLeNet (Acikgoz, 2022) (Acikgoz, 2022) CEEMDAN-AG-RE-EML (Acikgoz, 2022) 1h

Horizon 30 min 30 min 30 min 30 min 30 min 30 min 30 min 30 min 30 min 30 min 30 min 30 min 1 day 1 week 1 day 1 day 1h 1h 1h 2 day 1–12 hour 1h 1h 1h 1h 1h

34

Ekaterina A. Engel

AI Models for Maintenance of PV systems AI methods solve the most complex tasks, which include failure classification, detection, localization, and automated solar panel diagnostics, based on PV systems’ data (Figure4). Thus, grid operators can greatly increase the effectiveness and reliability of their solar plants based on AI methods. ANN, FL, DT, RNN, RF, and different ensembles automatically detected basic PV system faults based on data from ordinary sensors (Figure 4). DL and various types of CNN automatically perform analysis of infrared (IFR) images that are tracked by Unmanned Aerial Vehicles (UAVs). In this field of research, usually a dataset is highly unbalanced, i.e., it has unlabeled data and/or has rare failures. For this reason, the Balanced Accuracy, F1 score, Cohen’s Kappa, or MCC better reflect the model’s performance as compared to traditional accuracy metric. Most of the AI models were created based on the dataset which was generated from simulation. A limited number of failure classes were considered, with the exception of a number of works in (Bakdi et al., 2021; Liu et al., 2021) in which 10 or more faults were considered (Table 9). The Solar Plant Fault Forecasting Model on the Basis of a Modified Fuzzy Neural Net We proposed a failure forecasting model of a wiring losses’ failure free operating period of a PV box based on an MFNN that has two RNNs with fuzzy units (Engel, 2020). We created the MFNN based on a two-year historical dataset which included 20 kW PV array’s signals. We trained the MFNN based on the data � 𝑖𝑖 , 𝐼𝐼̃𝑖𝑖 �; 𝑍𝑍 𝑖𝑖 = �𝑇𝑇 𝑖𝑖 ; 𝑡𝑡 𝑖𝑖 ; 𝑉𝑉 𝑖𝑖 �; 𝑃𝑃𝑖𝑖 ) ( 𝑥𝑥 𝑖𝑖 = �𝑉𝑉� 𝑖𝑖 , 𝐼𝐼𝐼𝐼

(11)

𝑖𝑖 𝑖𝑖 where 𝑉𝑉� = ∫𝑓𝑓 𝑉𝑉 ; f is a time of the last PV box’s fault; 𝐼𝐼̃ = ∫𝑓𝑓 𝐼𝐼 ; 𝐼𝐼and 𝑉𝑉are the

� = ∫𝑖𝑖 𝐼𝐼𝐼𝐼; Ir is the irradiance; 𝑇𝑇 𝑖𝑖 and DC current and voltagerespectively; 𝐼𝐼𝐼𝐼 𝑓𝑓

𝑡𝑡 𝑖𝑖 are ambient and panel temperature respectively, 𝑃𝑃𝑖𝑖 – a wiring losses’ F-FOP of a PV box. Data (11) represents SCADA database of the Abakan solar plant PV boxes’ signals from 2015 till 2018. In this research the MFNN includes: a two-layer recurrent network (RNN), fuzzy units and two two-layered RNNs. We coded the MFNN architecture’s parameters (a number of hidden neurons – 𝑑𝑑𝑋𝑋,ℎ ∈ ����� 1. .7, corresponded weights and biases, number of RNN’s delays – 𝑑𝑑𝑋𝑋,𝑑𝑑 ∈ ����� 1. .3) into particles X. We coded the dimension component of particle X as 𝑑𝑑𝑋𝑋 =

Advances in the Artificial Intelligence Models for Photovoltaic Systems

35

�𝑑𝑑𝑋𝑋1 , 𝑑𝑑𝑋𝑋2 , 𝑑𝑑𝑋𝑋3 , 𝑑𝑑𝑋𝑋,𝑑𝑑1 , 𝑑𝑑𝑋𝑋,𝑑𝑑2 , 𝑑𝑑𝑋𝑋,𝑑𝑑3 � ∈ {𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = (1,1,1,1,1,1), 𝑑𝑑𝑚𝑚𝑚𝑚𝑚𝑚 = (7,7,7,3,3,3)}. The number of all𝑑𝑑𝑋𝑋 is 9261. We create the fault forecasting model in a solar plant on the basis of the created MFNN by elaborating the follows three steps. Step 1. The data (1) we classified into two groups accordingly with PV box’s mode: A1is normal (Ci=1), A2 is abnormal (Ci=0). We create two-layer RNN 𝑔𝑔(𝑍𝑍 𝑖𝑖 ). The vector 𝑍𝑍 and C are network’s input and target, respectively. We define fuzzy sets Aj, (A1is normal mode of PV box, A2 is abnormal) with membership function µj(s) base on two-layer RNN 𝑔𝑔(𝑍𝑍 𝑖𝑖 ), j=1. .2. Step 2. We create the MFNN which includes: the RNN 𝑔𝑔(𝑍𝑍 𝑖𝑖 ), and thenthe RNNs 𝑢𝑢𝑖𝑖 = 𝐹𝐹𝑗𝑗 �𝜇𝜇𝑗𝑗 �𝑍𝑍 𝑖𝑖 �, 𝑥𝑥 𝑖𝑖 �, 𝑗𝑗 = 1. .2. The MFNN architecture’s parameters (a 1. .7, corresponded weights and biases, number of hidden neurons – 𝑑𝑑𝑋𝑋,ℎ ∈ ����� ����� number of RNN’s delays – 𝑑𝑑𝑋𝑋,𝑑𝑑 ∈ 1. .3) have been coded into particles X. We coded the dimension component of particle X as 𝑑𝑑𝑋𝑋 = �𝑑𝑑𝑋𝑋1 , 𝑑𝑑𝑋𝑋2 , 𝑑𝑑𝑋𝑋3 , 𝑑𝑑𝑋𝑋,𝑑𝑑1 , 𝑑𝑑𝑋𝑋,𝑑𝑑2 , 𝑑𝑑𝑋𝑋,𝑑𝑑3 �. In this research the dimension of particle X is 𝐷𝐷(𝑑𝑑𝑋𝑋 ) = 3 + ∑3𝑞𝑞=1 𝑑𝑑𝑋𝑋,ℎ1𝑞𝑞 ∙ �𝑑𝑑𝑋𝑋,𝑑𝑑𝑞𝑞 + 4�. We define the fitness function as: 𝑓𝑓(𝐹𝐹) =

100% 𝐻𝐻

𝑖𝑖 𝑖𝑖 𝑖𝑖 ∙ ∑𝐻𝐻 𝑖𝑖=1�𝑃𝑃 − 𝑢𝑢 �/𝑃𝑃 ,

(12)

where H is number of evaluated samples. We tune MFFN 𝐹𝐹𝑗𝑗 based on the train 01/31/15. .11/30/18,and optimization algorithm o (if o=1 data (11), 𝑖𝑖 ∈ ���������������������������� then optimization algorithm is modified PSO (Forootan et al., 2022); if o=2 then optimization algorithm is proposed modified QPSO). The optimization algorithm o provides the best solution – the tuned MFNN 𝑦𝑦� 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑,𝑜𝑜 . If-then rules are defined as: Пj: IF 𝑍𝑍 𝑖𝑖 is Aj THEN 𝑢𝑢𝑖𝑖 = 𝐹𝐹𝑗𝑗 �𝜇𝜇𝑗𝑗 �𝑍𝑍 𝑖𝑖 �, 𝑥𝑥 𝑖𝑖 �. (13) Simulation of the tuned MFNN we briefly describe as follows. Step 1. Aggregation antecedents of the rules (12) maps input data 𝑍𝑍 𝑖𝑖 into their membership functions and matches data with conditions of rules. These mappings are then activating the k rule, which indicates the k PV box's operating mode and k agent’s subcultures – Sk, k =1. .2. Step 2. According the k mode the trained MFNN creates the forecasting F-FOP of a PV box 𝑢𝑢𝑖𝑖 = 𝐹𝐹𝑗𝑗 �𝜇𝜇𝑗𝑗 �𝑍𝑍 𝑖𝑖 �, 𝑥𝑥 𝑖𝑖 �. We fulfilled the solar plant fault-forecasting model based on the MFNN. We created the MFNN based on the training set of the data (11), 𝑖𝑖 ∈ ���������������������������� 01/31/15. .11/30/18. Due to achieve statistical results, we fulfilled 120 runs of modified PSO and proposed modified MD QPSO with parameters: 𝑆𝑆 =

36

Ekaterina A. Engel

150 (the number of particles are 150), 𝑇𝑇 = 450 (the maximum number of iterations is 450). Figure 1 11 shows that the modified MDQPSO has definitely more convergence speed than the modified PSO in the fault forecasting of the solar plant on the basis of the MFNN over training set of the data (10). The usage of the Nguyen-Widrow method to initialize swarm’s particles positions generates an effective MFNN’s architecture at initial iteration and accelerate a convergence of the modified QPSO. The proposed modified MDQPSO provides the best solution 𝑦𝑦� 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 which represents the tuned MFNN. This tuned MFNN includes the two-layer RNN 𝑔𝑔(𝑍𝑍 𝑖𝑖 ) with four hidden neurons (number of delays is two) and then the two two-layer RNN 𝐹𝐹𝑗𝑗 �𝜇𝜇𝑗𝑗 �𝑍𝑍 𝑖𝑖 �, 𝑥𝑥 𝑖𝑖 � with three and five hidden neurons, respectively (number of delays is one and two, respectively), 𝑗𝑗 = 1. .2. The experiments approved that the proposed QPSO with multi-swarms effectively tracked and fined the optimum dimension of the global minimum despite of height difference. 30 %

The modified PSO

25 %

The modified QPSO

The fitness function (2)

20 %

15 %

10 %

5%

0

0

50

100

150

200

Iteration No.

250

300

350

400

Figure 11. The mean convergence curves.

The simulation results from 120 runs showed that the developed modified QPSO prevents an early convergence to local minimum, exhibits the better performance, fast global convergence and calculation speed than does the case of the modified PSO. In forecasting an F-FOP of the solar plant the tuned MFFN reaches the 0.008% relative error (12) on a test set and outperforms RNN that reaches the 0.005% relative error (12) on a test set.

450

Advances in the Artificial Intelligence Models for Photovoltaic Systems

37

ML Models for Failure Diagnosis of the PV systems According to study (Arani and Hejazi, 2016), there are six different categories of PV systems failures: shading, open-circuit, degradation, line-to-line, bypass diode, and bridging. Frequent faults are failure in a component, system isolation, inverter shutdown, shading, and inverter MPP. In recent years, ML techniques that process data from ordinary sensors (Figure 4) have been highly applied for fault classification and, in some cases, to identify the location of a failure. In studies (Bakdi et al., 2021; Liu et al., 2021; Appiah et al., 2019; Chen et al., 2018; Pahwa et al., 2020; Hajji et al., 2021; Abbas and Zhang, 2021), researchers detect, classify, and localize (Abbas and Zhang, 2021) different failures of a PV system based on non-NN, ANN, ANFIS, and LSTM that simply process signals from ordinary sensors (Figure 4). In studies (Lu et al., 2019; Mustafa et al., 2022; Aziz et al., 2020; Gao and Wai, 2020; Chen et al., 2019; Hong and Pula, 2022; Wang et al., 2022), researchers detect, classify, and localize (Mustafa et al., 2022) different failures of a PV system based on CNNs. For this purpose, researchers tuned CNNs based on the created dataset which sample represented a twodimensional or three-dimensional transformation of data from ordinary sensors (Figure 4) namely, a scalogram (Aziz et al., 2020), a two-dimensional time series graph (Lu et al., 2019), a three-dimensional image (Chen et al., 2019) and a polar-coordinate image (Wang et al., 2022). This transformation can be simple (in Lu et al., 2019), only PV current and voltage were composed into a two-dimensional time series graph) or complex (in Chen et al., 2019), the direct current and alternating current values of a PV system were composed into a three-dimensional image based on a Gramian Angular Field; in (Wang et al., 2022), the time domain waveform signals were composed into a polar-coordinate image based on a symmetrized dot pattern (SDP). In Table 10, we summarize the ML models for PV failure diagnosis from studies.

38

Ekaterina A. Engel

Table 10. The Summary of the ML models for PV failure diagnosis Fault Diagnosis Stage Det

Clas

Loc

✓

✓

-

✓

✓

-

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Types of Faults Inverter fault, grid anomaly, mismatch fault, MPPT fault, converter fault Degradation, PS, PS w/BpD, short circuit, open, PS w/BpD short

Performance (%) False alarms < 1. Computational time is 11.809 s

98.3

line-to-line

97.66

hot spot

98.78

-

line-to-line, open circuit, degradation, and PS

99 accuracy that is superior as compared to DT

-

PS, bridging, bypass diode, temperature, short circuit, and complete shading

99.91 performance, which is superior as compared to DTs, XGBoost and RF

Healthy mode

98.17

inverter fault grid connection fault sensor fault panel fault panel connection fault

99.93 99.93 99.96 100.0

-

-

-

✓

Specific Data/Method (s) Applied/Ref. PCA-KDE-based multivariate KL divergence/ (Bakdi et al., 2021) Experimental data/ stacked autoencoder/ (Liu et al., 2021) Data with noise / (Arani, M.S.; Hejazi, 2016) LSTM/ (Appiah et al., 2019) Dataset that was created during simulation/ RF/ (Chen et al., 2018) Dataset with 1200 samples /ANNs/ (Pahwa et al., 2020) Dataset with 586,580 samples/PCA + RF/ (Hajji et al., 2021)

100.0

PS, open circuit, lineto-line, arc

70.45

Open-circuit, line-toline,

Average accuracy 99

PS w/ BpD, PS w/ reversed BpD, short circuit, increase series resistance

99.94 for Classification, 99.54 for Location

Scalograms with noise/CNN/ (Aziz et al., 2020) 2-D time series graph/CNN/ (Mustafa et al., 2022) CNN w/residual GRU/ (Mustafa et al., 2022)

Advances in the Artificial Intelligence Models for Photovoltaic Systems Fault Diagnosis Stage Det

Clas

Loc

✓

✓

✓

✓

✓

-

✓

✓

✓

✓

-

-

Types of Faults Line-to-line, opencircuit, short-circuit Short circuit, PS, abnormal aging, and hybrid failures

✓

R = 0.9989, RMSE = 0.0383 98.41

PS, degradation of a TF module, short circuit, open circuit

Average accuracy 95.78 which is superior as compared to CNN

Line-to-line

100.0

shorted modules in strings open module in strings shorted strings in arrays open strings in arrays healthy mode

Normal PV module ✓

Performance (%)

Specific Data/Method (s) Applied/Ref. ANFIS Sugeno/ (Abbas and Zhang, 2021) CNN and a fully connected module/ (Gao and Wai, 2020) Test dataset/ResNet/ (Chen et al., 2019) Dataset of 3D images/ 3D CNN/ (Hong and Pula, 2022)

91.67 91.67 100.0 95.24 100.0

100.0

✓ poor connection on a PV Module PV module breakage bypass diode

39

Dataset includes 3200 samples that generated by SDP, test dataset includes 800 samples (200 samples of each failure)/CNN/ (Wang et al., 2022)

100.0 100.0 99.5

Summarizing, we highlight a need for open datasets to assess experimental results on real test beds and an open tool to generate and process scalograms based on transformers with an attention mechanism which feasibly outperforms other ML methods, such as CNNs. For failure detection and classification, there is a need to study the MPPT algorithms based on Reinforcement Learning (RL) and a spiking neural network under failure conditions.

40

Ekaterina A. Engel

AI Models for Solar Panel Diagnostics The drop in PV system productivity due to deviant maintenance modes caused by nonclean module surfaces, cell damage, delamination, or hot spots, demands a solar panel diagnostic based on the AImodelsthatprocess the panels’ images. In studies (Starzyński et al., 2022; Manno et al., 2021; Pierdicca et al., 2020; Sizkouhi et al., 2021; Venkatesh and Sugumaran, 2022; Kurukuru et al., 2022; Zefri et al., 2022), researchers localized and identified different failures of a PV system based on CNNs that process the solar panels’ images, including thermographic images. In Table 11, we summarize the ML models for PV diagnostics from studies. Table 11. Comparison of ML Models for PV Diagnostics ML Method

YOLOv4 (Starzyński et al., 2022)

Localize/Identify Failure Light reflex, hot spot, short circuit, faulty string/sunbstring, “good” module

Performance 0.96 0.956 0.905 0.969 0.997 Average performance on test dataset is 98%, a range of processing speed is [0.001, 2 min]

Dataset Preprocessed dataset of thermographic images Preprocessed dataset of thermographic images Dataset of thermal images generated by infrared sensors installed in a UAV Dataset of 1000 affected images

CNN (Manno et al., 2021)

Binary classification of hot spots

Hybrid mask region CNN (Pierdicca et al., 2020)

classify three failures: one damaged cell, nonadjacent, and adjacent damaged cells

RMSE of 26.85 W/m2, 19.78 W/m2 and 39.2 W/m2 for PC, MC, and TF PV systems correspondingly

detect a failure (bird’s drops over a PV array)

Average performance on test dataset is 93%

Delamination, hot spot, glass damages, decolorization, and snail trails

Best accuracy is 100%

Dataset of aerial images.

The average accuracy on test dataset is 94.4%

The thermographic images dataset that includes 16,000 samples (1600 for each class)

Modified VGG16 (Sizkouhi et al., 2021) SVM, naive Bayes, kNN, DT, RF and pre-trained DNN (Venkatesh and Sugumaran, 2022) DIP filters and SVM classifier (Kurukuru et al., 2022)

Classification into 10 different classes (1 healthy and 9 failure modes including warm module/substrings/cells, hot spot, etc.)

Advances in the Artificial Intelligence Models for Photovoltaic Systems ML Method

VGG16 (Zefri et al., 2022)

Localize/Identify Failure Localization and classification into 6 different classes (1 healthy and 5 failure modes including overheated module/substrings hot spot, etc.)

41

Performance

Dataset

The mean F1-score is 94.52%

Dataset of thermal infrared images was collected from 28 PV systems, which have 93220 solar panels

Summarizing, we identify an opportunity to collect and make datasets available in which new AI models for PV system diagnostics can be tested. In the reviewed studies, a considerable number of ML models process images almost perfectly. In the reviewed studies, the ML models where signals of image sensor and the CNN blocks strongly correlate provide high performance. There is an argumentative direction to substitute non-NN smart models with a DNN-based model for the PV system’s maintenance because DNN provides better information processing quality and performance as compared to non-NN smart models. In addition, GANs can be applied to generate artificial thermal images and create knowledge of the failure. Moreover, future research can comprise the elaboration of a pipeline for implementing a real time PV array diagnostic system based on DNN or spiking neural network.

AI Models for Control of PV Systems The application of AI methods for the MPPT of PV systems has massive potential to increase their stability, reliability, dynamic response, and other essential advancements and easing their integration to electric grids. AI MPPT Models of PV Systems The insolation and cell temperature of solar panels primarily define the total generated power by a PV system. In the research reviews, a whole array of differing MPPT algorithms has been revealed (Kurukuru et al., 2021; Forootan et al., 2022; Massaro and Starace, 2022). Among them, the perturbation and observation (P and O) and incremental conductance (INC) algorithms are the most popular due to their easy and simple implementation. However, controllers which were created on the basis of these algorithms for PV systems have very bad speed of the response times, a long time to settle down from oscillating around the reference state. Furthermore, under PS, the MPPT task

42

Ekaterina A. Engel

demands GO. Thus, traditional algorithms for MPPT do not provide global MPPT (GMPPT) and decrease efficiency in solar power production. There are a lot of GO algorithms to create a GMPPT model (Kurukuru et al., 2021; Forootan et al., 2022; Massaro and Starace, 2022), but all these models have the following disadvantages: power oscillations in the calm mode; the initialization is a critical issue that decrease power; very slow convergence to a GMPP under insolation’s variation, etc. Due to all the abovementioned disadvantages, GO-based, real-time GMPPT of a PV system are ineffective while AI models provide the required performance. In Table 12, we summarize the AI models for MPPT of a PV system from studies. In (Bag and Ray, 2019), researchers integrated the trained RL control agent into a fuzzy-logic-sliding mode control and incremental conductancesliding mode control (RL FL INC) and gained better performance as compared to a classical RL agent (Table 12). In (Bouarroudj et al., 2019), researchers created an MPPT controller based on a fuzzy logic search of variable voltage step size and fuzzy adaptive RBFNN. The simulation results reflect the superiority of the developed MPPT controller as compared to the conventional P and O and RBF-NN. In (Engel and Engel, 2020), we introduced the GMPPT system based on an MFNN that has five convolutional blocks to process the PV array’s images, RNNs, and fuzzy units. Figure 12 shows the proposed GMPPT system based on an MFNN, where 𝐼𝐼𝑚𝑚𝑖𝑖 is image of PV modules; 𝑥𝑥 𝑖𝑖 = ( 𝑉𝑉 𝑖𝑖 , 𝑃𝑃𝑖𝑖−1 , 𝑑𝑑𝑑𝑑/ 𝑑𝑑𝑉𝑉 𝑖𝑖 ) and 𝑢𝑢𝑖𝑖 – input and output signal of MFNN, correspondingly; 𝜇𝜇𝑗𝑗 —membership function of the fuzzy sets 𝐴𝐴𝑗𝑗 (𝐴𝐴1 is the rapidly increased uniform insolation, 𝐴𝐴2 is nonuniform insolation); z = ind max(𝜇𝜇𝑗𝑗 ) = {j |∀𝑘𝑘 ≠ 𝑗𝑗𝜇𝜇𝑗𝑗 ≥ 𝜇𝜇𝑘𝑘 } triggers 𝑗𝑗

the rule, which corresponds the 𝑧𝑧 fuzzy set andRNN 𝐹𝐹𝑧𝑧 .The performance and control speed in GMPPT under PS of the created MFNN were superior as compared to the PSO and RNNs.

Figure 12. The GMPPT system based on an MFNN.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

43

Table 12. Recent comparative studies of AI-based MPPT implementations AI Method for MPPT RL control agent (Kofinas et al., 2017) RL FL INC (Bag and Ray, 2019) Q-learning (Bavarinos et al., 2021) Q-learning (Kalogerakis et al., 2020) Q-learning (DQL) agent (Phan et al., 2020) Deep deterministic policy gradient MPPT (Lapan, 2018) Q-table MPPT (Chou et al., 2019) Fuzzy Adaptive RBF-NN (Bouarroudj et al., 2019) MFNN (Engel and Engel, 2020) DL RL agent (Avila et al., 2020) Bayesian ML (BML) (Keyrouz, 2018) ANN (Behera, 2018) Feedback Linearization (FBL) embedded Full Recurrent Adaptive NeuroFuzzy (FRANF) (Awais et al., 2020) Hermite Wavelet-embedded Neural Fuzzy (Hassan et al., 2017)

Software Platform

MPPT Simulation Time (s)

Steady-State Oscillation (%)

MPPT Efficiency (%)

Simulink

-

Almost zero

99.4

1

-

99.8

20

-

-

MATLAB/Simu link MATLAB and Simulink R2015a MATLAB/Simu link MATLAB/Simu link

30

98.97

8

±2

97

40

-

97.5

40

-

97.5

1.5

Almost zero

99.21

8

Almost zero

99.3

10

-

99

30

Almost zero

98.9

10

-

99

MATLAB/Simu link

25

-

90.2

MATLAB/Simu link

12

MATLAB Simulink

and

MATLAB and Simulink R2017b MATLAB/Simu link Authors’ software OpenAI Gym environment MATLAB 2013a/Simulink MATLAB/Simu link

94.04

Summarizing, we highlight a demand for implementing more benchmarks for performance comparison of the real-time MPPT AI models based on ML frameworks, which we presented in earlier Section. In addition, a real-time MPPT model based on a spiking neural network is a promising alternative.

44

Ekaterina A. Engel

AI Models for Control of Reconfigurable PV systems The technology of reconfigurable PV arrays (rPV) by switching the electrical interconnection maximizes the generated PV array power in case of PS (Baka et al., 2019; Lakshika et al, 2020). There are two classes of rPV: static and dynamic. Researchers proposed a lot of rPV’s structures, including Honey Comb, Series Parallel, Total Cross Tied (TCT), etc. (Baka et al., 2019; Lakshika et al, 2020; Deshkar et al., 2015; Osmani et al., 2022; Ajmal et al., 2021; Ibrahim et al., 2021; Belhachat and Larbes, 2017; Solis-Cisneros al., 2022; Nguyen-Duc et al., 2022; Engel et al., 2021; Natsheh and Samara, 2020; Warden and Situnayake, 2021; Sairam et al., 2022). According to the articles (Sama, 2021; Uehara et al., 2021; Vandemark, 2020) the last one generates more power in case of PS as compared to other structures. The GMMPT of an rPV array in case of PS represents a GO task. In Table 13, we summarize the AI models for rPV from these studies. The comparative analysis of recent rPV methods in (Osmani et al., 2022) revealed that a TCT rPV based on a Static Shade Dispersion Physical Array Relocation (SD-PAR) algorithm and Modified Harris Hawks Optimizer (MHHO) algorithm that generated a switching matrix generates more power under PS as compared to other algorithms. Although, all metaheuristic optimizers do not provide a GMMP in real time mode because of a slow convergence. The goal of study (Engel et al., 2021) is a GMPPT of an rPV array based on the MFNN in a case of PS. We created an optimal MFNN based on the dataset that contains the 20 kW PV array’s signals under PS including PV array images that were congregated at the town Abakan from 31 January 2018 through 31 December 2018. Figure 13 and Figure 14 display the insolation of the four solar panels’ groups for the time period 9:20 am 3 December 2018– 9:21 am 3 December 2018. Figure 15 shows that the rPV system based on the MFNN outperforms an rPV system based on GA because last one does not provide GMPP in this case. Similarly, we evaluated the performances of the rPV system based on the MFNN and rPV system based on GA on 100 test samples from the time period 1 December 2018–31 December 2018. The comparative simulation results show the superiority in terms of robustness and control speed of the created intelligent rPV system under PS that provides on average 30% more energy as compared to a TCT rPV system based on GA.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

45

Table 13. Recent comparative studies of AI models for rPV AI Technology Advantages/Disadvantages TCT rPV based on Static Shade Dispersion Physical Array Relocation (SD-PAR) algorithm and Modified Harris Hawks Optimizer (MHHO) (Osmani et al., 2022)

Performance AI Technology

Advantages/Disadvantages

Disadvantage: GO-based, realtime GMPPT of a PV system are ineffective because of the slow convergence

Technology generates more power under PS as compared to other algorithms.

Reconfiguration algorithms based on a GA (Deshkar et al., 2015, Ajmal et al., 2021)

Disadvantage: GO-based, realtime GMPPT of a PV system are ineffective because of the slow convergence

ANFIS and an OCS (Ibrahim et al., 2021, Belhachat and Larbes, 2017)

-

Fuzzy controller Cisneros al., 2022)

(Solis-

CNNs (Nguyen-Duc et al., 2022)

MFNN that contains: a convolutional block, RNNs and fuzzy units (Engel et al., 2021)

Disadvantage: the proposed scheme does not provide MPPT under dynamic PS due to constant threshold-based switching of a fuzzy controller. Advantage: Eight CNNs are implemented by PyTorch and validated on 1842 images under four PS scenarios Advantage: MFNN is implemented by authors software. The trained MFNN by processing of the signals from ordinary sensors and PV array’s image creates the GMMP interconnection matrix and GMMP voltage in case of PS.

The simulation results in Simulink for TCT rPV revealed that the developed algorithm increased power: by 16.68% and 6.8% in three PS scenarios as compared to the TCT and the Su Do Ku scheme [129]; in four PS scenarios as compared to TCT. Created algorithm provided faster GMMPT and an average of 21% more generated power as compared to the P and O algorithm

-

The VGG 19 provides the best result (MAPE is 3.75%, RMSE is 0.0513, accuracy is 88.47%). The results show the superiority of the created intelligent rPV system under PS in terms of robustness, control speed that provides on average 30% more energy, as compared to a TCT rPV system based on GA

46

Ekaterina A. Engel

Figure 13. The configuration scheme created by (a) the TCT rPV, (b) the rPV system based on the MFNN, and (c) the rPV system based on GA. The insolation level of the first PV modules group The insolation level of the second PV modules group The insolation level of the third PV modules group

700

The insolation level of the fourth PV modules group

Insolation (Wt/m

2

)

600

500

400

300

200 0

0.5

Time (min.)

Figure 14. The solar panels groups’ insolation.

1

Advances in the Artificial Intelligence Models for Photovoltaic Systems

47

Power (kW)

8

7

The reconfigurable TCT PV system based on GA Maximum power of the PV array The intelligent reconfigurable TCT PV system

6 0

0.5

1

Time (Min.)

Figure 15. Curves of the generated power of rPV system based on the MFNN and GA.

Summarizing, we identify an opportunity to use RNN for rPV that provides a GMMP interconnection matrix and GMMP voltage under dynamic PS. Nevertheless, an rPV’s payback period is about 20 years (Baka et al., 2019) solely in places where PS happens daily, or over the full year leastwise in the seasons where solar production is great.

Future Technologies for Smart Solar Energy The long-term contribution, including increased capacity of solar energy, depends on solving the remaining tasks of gridsintegration, high costs, and low efficiency, mainly through the research and development of a smart PV system based on integration of cutting-edge technologies, including DNN (Natsheh and Samara, 2020; Warden and Situnayake, 2021; Sairam et al., 2022; Sama, 2021; Uehara et al., 2021; Vandemark, 2020; Wang et al., 2020; Matsuo et al., 2022; Mellit et al., 2022; Tina et al., 2021; Mateo Romero et al., 2022; Alipour et al., 2014; Baran, 2014. To attain the smart optimization and high efficiency of solar energy, the cloud, big data, AI, EC, IoT, quantum, and sensor technologies need to be adaptively combined. Figure 16 reflects the overlapping integration of these technologies into a smart PV system. The integration of the above-mentioned cutting-edge technologies provides high efficiency of AImodels for the PV

48

Ekaterina A. Engel

system’s design, forecasting, maintenance, and control. Implementation of such cutting-edge AI technologies for the PV system’s design, forecasting, maintenance, and control provides digital transformation of solar energy into smart solar energy. These AI models are implementations of a basic ML model class which is represented on Figure 16 as an UML class diagram. Figure 16 shows a method “Add” of a basic ML model class. This method adds a quantum layer into a classical ML model to create a quantum ML model. This method can be implemented by an integrating framework (Pennylane) for quantum computer simulators (Sama, 2021). A quantumbased PV system failure detection model was developed in Uehara et al., 2021. IoT provides an optimal solution to collect solar energy big data wirelessly (Figure 16). In (Natsheh and Samara, 2020), the solution researchers integrated a PV system failure detection AI model.

Figure 16. Smart PV system.

Future research can comprise the elaboration of a pipeline for implementing a real time PV array diagnostic system based on IoT, EC, and/or

Advances in the Artificial Intelligence Models for Photovoltaic Systems

49

TinyML technologies (Warden and Situnayake, 2021). In (Sairam et al., 2022), researchers developed based on EC a lightweight AI real-time PV system failure detection model. Recently in (Uehara et al., 2021; Vandemark, 2020) researchers developed the cloud-based monitoring solutions. The most complex issue of a ML model is self-learning. The potential methods for ML model’s self-learning are memristors and a spiking neural network (Wang et al., 2020). In the future, a smart PV system will integrate a self-supervised learning AI model zoo (Matsuo et al., 2022) that provides optimum AI models for the PV system’s design, forecasting, maintenance, and control. The integration of the cloud, big data, AI, EC, IoT, quantum, and sensor technologies will provide high efficiency of AI models for the PV system’s design, forecasting, maintenance, and control. Implementation of these models for the PV system’s design, forecasting, maintenance, and control provides digital transformation of solar energy into smart solar energy. The integrated electric grids are becoming increasingly reliable and overall solar production costs are minimized.

Discussion In the reviewed studies, the AI models where signals of image sensor and the CNN blocks strongly correlate provide high performance. There is an argumentative direction to substitute non-NN smart models with a DNN-based model for the PV system’s design, forecasting, maintenance, and control because DNN provides better information processing quality and performance as compared to non-NN smart models. The impact of the AI models based on the proposed implementation scheme on a PV system value chain will mostly be associated with the cost of software development, which implements a ML PV system based on ONNX, a developed software’s integration with SCADA, and maintenance. • •

The most complex issue of a AI PV system is self-learning of a ML model. The potential methods for adaptive learning are memristors and a spiking neural network. In addition, we have outlined several problems that can be considered for future research in field of smart solar energy:

50

Ekaterina A. Engel

• • • • •

In forecasting and failure detection, the usage of the DNNs such as transformers based on an attention mechanism is a promising alternative. For failure detection and classification, there is a need to study the MPPT algorithms based on RL and a spiking neural network under failure conditions. For diagnosis of a PVarray based on thermal images, the usage of GANs is a promising alternative. There is a need to propose a pipeline for implementing a real-time PVarray diagnostic system based on IoT, EC, and/or TinyML technologies. The development of AI algorithms for real-time processing and decision making are most in demand in PV systems.

Within the EU COVID-19 strategic reply, the smart energy standards define a cloud platform specification for a distributed smart solar big data ecosystem that will provide the creation of effective AImodels for smart solar energy. The open solar energy data sources, including big data, provide the development of cutting-edge AI technologies in solar energy. Therefore, more open datasets with real data from PV systems should be shared with the research community. In order to achieve the smart optimization and high efficiency of solar energy, the cloud, big data, AI, EC, IoT, quantum, and sensor technologies need to be adaptively combined and implemented as smart grid, home, and city applications. The integration of the above-mentioned cutting-edge technologies will provide high efficiency of AI models for the PV system’s design, forecasting, maintenance, and control. Implementation of these models for the PV system’s design, forecasting, maintenance, and control will provide digital transformation of solar energy into smart solar energy. The integrated electric grids are becoming increasingly reliable, and overall solar production costs are minimized. Forthcoming AI technologies for solar energy will integrate cloud-based solutions, in which these technologies take full benefits of the ML parallelism, data parallelism, practically limitless big data and AI knowledge storage, and almost boundless parallel computational resources.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

51

Conclusions We presented a structured (mostly in benchmark tables) study of the advances in AI models for the PV system’s design, forecasting, maintenance, and control where most of the reviewed articles were published within the last five years. AI models are key elements of PV systems because they automatically create smart models for the PV system’s design, forecasting, maintenance, and control and more effectively analyze exponentially growing big data as compared to traditional algorithms. In this review, we briefly summarized our self-adaptive models for sizing, forecasting, maintenance, and control of a PV system based on an MFNN that were automatically created with regard to a task’s complexity and overfitting problem. The long-term contribution, including increased capacity of solar energy, depends on solving the remaining tasks of coupling to electric grids, high costs, and low efficiency, mainly through the research and development of a smart PV system based on the integration of cutting-edge technologies, including DNN.

Disclaimer None

Acknowledgments The study was funded by a grant from the Ministry of Education and Science of the Republic of Khakassia (Agreement No. 91 dated 12/13/22). The reported study was fulfilledaccording the activity “Development of intelligent systems for forecasting and maximizing power generation based on the original modified fuzzy neural network, their implementation as software and the implementation at a renewable power plant” within program of the World-class Scientific Educational Center “Yenisei Siberia”.

52

Ekaterina A. Engel

References Abbas M, Zhang D. A smart fault detection approach for PV modules using adaptive neurofuzzy inference framework. Energy Rep. (2021) 7: 2962-2975. Abdellatif A, Mubarak H, Ahmad S, Ahmed T, Shafiullah GM, Hammoudeh A, Abdellatef H, Rahman MM, Gheni HM. Forecasting Photovoltaic Power Generation with a Stacking Ensemble Model. Sustainability (2022) 14: 11083. Abdellatif H, Hossain MI, Abido MA. Parameters Estimation of PV Models Using Artificial Neural Network. Arab J Sci Eng (2022) 1-10. Abdel-Nasser M, Mahmoud K. Accurate photovoltaic power forecasting models using deep LSTM-RNN. Neural Comput. Appl. (2019) 31: 2727–2740. Acikgoz H. A novel approach based on integration of convolutional neural networks and deep feature selection for short-term solar radiation forecasting. Applied Energy (2022) 305: 117912. Ahn HK, Park N. Deep RNN-Based Photovoltaic Power Short-Term Forecast Using Power IoT Sensors. Energies (2021) 14: 436. Ajmal AM, Ramachandaramurthy VK, Tomar A, Ekanayake JB. Optimal dynamic reconfiguration of large-scale PV plant under partial shading conditions based on two reconfigurable stages. Int Trans Electr Energ Syst. (2021) 31: 12746. Akhter MN, Mekhilef S, Mokhlis H, Almohaimeed ZM, Muhammad MA, Khairuddin AS, Akram R, Hussain MM. An Hour-Ahead PV Power Forecasting Method Based on an RNN-LSTM Model for Three Different PV Plants. Energies (2022) 15: 2243 Akhter MN, Mekhilef S, Mokhlis H, Mohamed Shah N. Review on forecasting of photovoltaic power generation based on machine learning and metaheuristic techniques. IET Renew. Power Gener. (2019) 13: 1009–1023. Alipour S, Jannat F, Hosseini L. Teaching Breast Cancer Screening via Text Messages as Part of Continuing Education for Working Nurses: A Case-control Study. Asian Pac J Cancer P (2014) 15(14): 5607-5609. Alkhayat G, Mehmood R. A review and taxonomy of wind and solar energy forecasting methods based on deep learning. Energy AI (2021) 4:100060 Appiah AY, Zhang X, Ayawli BB, Kyeremeh F. Long Short-Term Memory Networks Based Automatic Feature Extraction for Photovoltaic Array Fault Diagnosis. IEEE Access (2019) 7:30089-30101. Arani MS, Hejazi MA. The Comprehensive Study of Electrical Faults in PV Arrays. J. Electr. Comput. Eng. (2016) 8712960. Avila L, De Paula M, Trimboli M, Carlucho I. Deep reinforcement learning approach for MPPT control of partially shaded PV systems in smart grids. Applied Soft Computing (2020) 97: 106711. Avila L, Paula MD, Trimboli M, Carlucho I. loavila/mppt-gym. Available online: https://github.com/loavila/mppt-gym (11.09.2022). Awadallah, M. Parameters estimation of photovoltaic modules: Comparison of ANFIS and ANN. International Journal of Industrial Electronics and Drives (2014) 1: 121-129.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

53

Awais M, Khan L, Ahmad S, Mumtaz S, Badar R. Nonlinear adaptive NeuroFuzzy feedback linearization based MPPT control schemes for photovoltaic system in microgrid. PLoS ONE (2020) 15: 0234992. Bag A, Subudhi B, Ray PK. A combined reinforcement learning and sliding mode control scheme for grid integration of a PV system. CSEE Journal of Power and Energy Systems (2019) 5: 498-506. Baka M, Manganiello P, Soudris D, Catthoor F. A cost-benefit analysis for reconfigurable PV modules under shading, Solar Energy (2019) 178: 69–78. Bakdi A, Bounoua W, Guichi A, Mekhilef S. Real-time fault detection in PV systems under MPPT using PMU and high-frequency multi-sensor data through online PCA-KDEbased multivariate KL divergence. Int. J. Electr. Power Energy Syst. (2021) 125: 106457. Baran E. A Review of Research on Mobile Learning in Teacher Education. Educational Technology & Society (2014) 17(4): 17-32. Bavarinos K, Dounis A, Kofinas P. Maximum Power Point Tracking Based on Reinforcement Learning Using Evolutionary Optimization Algorithms. Energies (2021) 14:335. Behera MK, Majumder I, Nayak N. Solar photovoltaic power forecasting using optimized modified extreme learning machine technique. Engineering Science and Technology (2018) 21: 428-438 Belhachat F, Larbes C. Global maximum power point tracking based on ANFIS approach for PV array configurations under partial shading conditions. Renew. Sustain. Energy Rev. (2017) 77:875-889. Belmahdi B, Louzazni M, El Bouardi A. Comparative optimization of global solar radiation forecasting using machine learning and time series models. Environmental Science and Pollution Research (2022) 29: 14871-14888. Berghout T, Benbouzid M, Bentrcia T, Ma X, Djurović S, Mouss LH. Machine LearningBased Condition Monitoring for PV Systems: State of the Art and Future Prospects. Energies (2021) 14:6316. Bouarroudj N, Boukhetala D, Feliu-Batlle V, Boudjema F, Benlahbib B, Batoun B. Maximum power point tracker based on fuzzy adaptive radial basis function neural network for PV-system. Energies (2019) 12: 2827. Bradbury K, Saboo R, L Johnson T, Malof JM, Devarajan A, Zhang W, M Collins L, G Newell R. Distributed Solar Photovoltaic Array Location and Extent Data Set for Remote Sensing Object Identification. Scientific data (2016)3(1): 1-9. Chen X, Yu K. Hybridizing cuckoo search algorithm with biogeography-based optimization for estimating photovoltaic model parameters. Solar Energy (2019) 180:192–206. Chen Z, Chen Y, Wu L, Cheng S, Lin P. Deep residual network based fault detection and diagnosis of photovoltaic arrays using current–voltage curves and ambient conditions. Energy Convers. Manage. (2019) 198.

54

Ekaterina A. Engel

Chen Z, Han F, Wu L, Yu J, Cheng S, Lin P, Chen H. Random Forest based intelligent fault diagnosis for PV arrays using array voltage and string currents. Energy Conversion and Management (2018) 178: 250-264. Chou KY, Yang ST, Chen YP. Maximum Power Point Tracking of Photovoltaic System Based on Reinforcement Learning. Sensors (2019) 19: 5054. Das UK, Tey KS, Seyedmahmoudian M, Mekhilef S, Idris MY, Van Deventer W, Horan B, Stojcevski A. Forecasting of photovoltaic power generation and model optimization: A review. Renew. Sustain. Energy Rev. (2018) 81: 912–928. Degtyarev AS, Kosenko VI, Engel EA, Engel NN, Savelyeva MV. Power forecasting for a photovoltaic system based on the multi-agent adaptive fuzzy neuronet. IOP Conf. Series: Materials Science and Engineering (2018) 450:072012. Deshkar SN, Dhale SB, Mukherjee JS, Babu TS, Rajasekar. Solar PV array reconfiguration under partial shading conditions for maximum power extraction using genetic algorithm. Renewable and Sustainable Energy Reviews (2015) 43: 102-110. Diab AZ, Sultan HM, Aljendy R, Al-Sumaiti A, Shoyama M, Ali ZM. Tree growth-based optimization algorithm for parameter extraction of different models of photovoltaic cells and modules. IEEE Access (2020) 5:119668–119687. DKASC, Alice springs | DKA solar centre. Available online: http://dkasolarcentre.com.au/ locations/alice-springs (12.09.2022). Ebrahimi SM, Salahshour E, Malekzadeh M, Gordillo F. Parameters identification of PV solar cells and modules using flexible particle swarm optimization algorithm. Energy (2019) 179: 358–372. EITCI Institute | Smart Energy Standards Group. Available online: https://eitci.org/sesg (10.09.2022). Elazab OS, Hasanien HM, Elgendy MA, Abdeen AM. Whale optimization algorithm for photovoltaic model identification. The Journal of Engineering (2017) 13:1906–1911. Engel E, Engel N. Automatic adaptation of a Solar Plant Intelligent Control System. Journal of Physics: Conference Series (2021) 1889. Engel E, Engel N. A Review on Machine Learning Applications for Solar Plants. Sensors (2022) 22: 9060. Engel E, Engel N. Photovoltaic System Control Model on the Basis of a Modified Fuzzy Neural Net. Studies in Computational Intelligence (2020) 856: 45–52. Engel EA, Engel NE, Kovalev I, Testoyedov N. Intelligent reconfigurable photovoltaic system. Energies (2021) 14: 7969. Engel EA, Engel NE. Maximum photovoltaic array power point tracking algorithm based on modified particle swarm optimization under non-uniform irradiances. IOP Conf. Series: Materials Science and Engineering (2020) 734: 012112. Engel EA, Engel NE. Solar Plant Intelligent Control System Under Uniform and Nonuniform Insolation Advances in Neural Computation, Machine Learning, and Cognitive Research IV. Neuroinformatics 2020. Studies in Computational Intelligence (2021) 925: 374-380.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

55

Engel EА. Photovoltaic Applications on the Basis of Modified Fuzzy Neural Net Solar Irradiance. Types and Applications, In Nova Science Publishers, ed Daryl M. Welsh (2020) 7–87. ESOLMET-IER instituto de energíasrenovables. Available online: http://esolmet.ier.unam. mx/Tipos_consulta.php (12.09.2022). F. Aziz, A. UlHaq, S. Ahmad, Y. Mahmoud, M. Jalal and U. Ali. A Novel Convolutional Neural Network-Based Approach for Fault Classification in Photovoltaic Arrays. IEEE Access (2020) 8:41889-41904. Faisal AF, Rahman A, Habib MT, Siddique AH, Hasan M, Khan MM. Neural networks based multivariate time series forecasting of solar radiation using meteorological data of different cities of Bangladesh. Results in Engineering (2022) 13: 100365. Forootan MM, Larki I, Zahedi R, Ahmadi A. Machine Learning and Deep Learning in Energy Systems: A Review. Sustainability (2022) 14:4832. Gao W, Wai RJ. A Novel Fault Identification Method for Photovoltaic Array via Convolutional Neural Network and Residual Gated Recurrent Unit. IEEE Access (2020) 8:159493-159510. Gaviria JF, Narváez G, Guillen C, Giraldo LF, Bressan M. Machine learning in photovoltaic systems: A review. Renewable Energy (2022) 196:298-318. GNU Octave 7.2.0. Available online: https://octave.org/download (15.10.2022) Grimaccia F, Leva S, Mussetta M, Ogliari E. ANN Sizing Procedure for the Day-Ahead Output Power Forecast of a PV Plant. Appl. Sci. (2017) 7: 622. Guermoui M, Benkaciali S, Gairaa K, Bouchouicha K, Boulmaiz T, Boland JW. A novel ensemble learning approach for hourly global solar radiation forecasting. Neural Computing and Applications (2022) 34: 2983-3005. Ha N, Xu K, Ren G, Mitchell A, Ou JZ. Machine Learning-Enabled Smart Sensor Systems. Adv. Intell. Syst. (2020) 2:2000063. Hajji M, Harkat MF, Kouadri A, Abodayeh K, Mansouri M, Nounou H, Nounou M. Multivariate feature extraction based supervised machine learning for fault detection and diagnosis in photovoltaic systems. European Journal of Control (2021) 59: 313321. Hassan SZ, Li H, Kamal T, Arifoğlu U, Mumtaz S, Khan L. Neuro-Fuzzy Wavelet Based Adaptive MPPT Algorithm for Photovoltaic Systems. Energies (2017) 10: 394. Hong YY, Pula RA. Detection and classification of faults in photovoltaic arrays using a 3D convolutional neural network. Energy (2022) 246: 123391. Hossain M, Mekhilef S, Danesh M, Olatomiwa L, Shamshirband S. Application of extreme learning machine for short term output power forecasting of three grid-connected PV systems. J. Clean. Prod. (2017) 167:395–405. Hossain MS, Mahmood H. Short-Term Photovoltaic Power Forecasting Using an LSTM Neural Network and Synthetic Weather Forecast. IEEE Access (2020) 8: 172524172533.

56

Ekaterina A. Engel

Ibrahim SA, Nasr A, Enany MA. Maximum Power Point Tracking Using ANFIS for a Reconfigurable PV-Based Battery Charger Under Non-Uniform Operating Conditions. IEEE Access (2021) 9:114457-114467. Kalogerakis C, Koutroulis E, Lagoudakis MG. Global MPPT Based on Machine-Learning for PV Arrays Operating under Partial Shading Conditions. Appl. Sci. (2020) 10: 700. Kang T, Yao J, Yang S, Duong T L, Zhu X. Novel cuckoo search algorithm with quasioppositional population initialization strategy for solar cell parameters identification. 13th World Congress on Intelligent Control and Automation (2018) 4-8. Keyrouz F. Enhanced Bayesian based MPPT controller for PV systems. IEEE Power and Energy Technology Systems Journal (2018) 5: 11-17. Khan A, Bhatnagar R, Masrani V, Lobo VB. A Comparative Study on Solar Power Forecasting using Ensemble Learning. In proceeding of 4th international conference on trends in electronics and informatics (ICOEI). (2020) 224-231. Khan PW, Byun YC, Lee SJ, Kang DH, Kang JY, Park HS. Machine Learning-Based Approach to Predict Energy Consumption of Renewable and Nonrenewable Power Sources. Energies (2020) 13: 4870. Khatib T, Elmenreich W. An Improved Method for Sizing Standalone Photovoltaic Systems Using Generalized Regression Neural Network. International Journal of Photoenergy (2014) 8. Kofinas P, Doltsinis S, Dounis AI, Vouros GA. A reinforcement learning approach for MPPT control method of photovoltaic sources. Renewable Energy (2017) 108: 461473. Kumari P, Toshniwal D. Extreme gradient boosting and deep neural network based ensemble learning approach to forecast hourly solar irradiance. J. Clean. Prod. (2021) 279: 123285. Kurukuru VB, Haque A, Tripathy AK, Khan MA. Machine learning framework for photovoltaic module defect detection with infrared images. Int. J. Syst. Assur. Eng. Manag. (2022) 1–17. Kurukuru VS, Haque A, Khan MA, Sahoo S, Malik A, Blaabjerg F. A Review on Artificial Intelligence Applications for Grid-Connected Solar Photovoltaic Systems. Energies (2021) 14:4690. Lakshika KH, Boralessa MK, Perera MK, Wadduwage DP, Saravanan V, Hemapala KM. Reconfigurable solar photovoltaic systems: A review. Heliyon (2020) 6(11): e05530. Lapan M. Deep Reinforcement Learning Hands-On: Apply modern RL methods, with deep Q-networks, value iteration, policy gradients, TRPO, AlphaGo Zero and more. Packt Publishing Ltd. (2018). Li S, Gong W, Yan X, Hu C, Bai D, Wang L. Parameter estimation of photovoltaic models with memetic adaptive differential evolution. Solar Energy (2019) 190: 465–474. Lin P, Cheng S, Yeh W, Chen Z, Wu L. Parameters extraction of solar cell models using a modified simplified swarm optimization algorithm. Solar Energy (2017) 144:594– 603.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

57

Liu L, Liu D, Sun Q, Li H, Wennersten R. Forecasting Power Output of Photovoltaic System Using A BP Network Method. Energy Procedia (2017) 142: 780–786. Liu Y, Ding K, Zhang J, Li Y, Yang Z, Zheng W, Chen X. Fault diagnosis approach for photovoltaic array based on the stacked auto-encoder and clustering with I-V curves. Energy Convers. Manag. (2021) 245: 114603. Lu X, Lin P, Cheng S, Lin Y, Chen Z, Wu L, Zheng Q. Fault diagnosis for photovoltaic array based on convolutional neural network and electrical time series graph. Energy Conversion and Management (2019) 196: 950-965. Luo X, Zhang D, Zhu X. Combining transfer learning and constrained long short-term memory for power generation forecasting of newly-constructed photovoltaic plants. Renewable Energy (2022) 185: 1062-1077. Malof JM, Li B, Huang B, Bradbury K, Stretslov A. Mapping Solar Array Location, Size, and Capacity Using Deep Learning and Overhead Imagery. ArXiv:1902.10895 (2019) 6. Manno D, Cipriani G, Ciulla G, Di Dio V, Guarino S, Brano VL. Deep learning strategies for automatic fault diagnosis in photovoltaic systems by thermographic images. Energy Convers. Manag. (2021) 241:114315. Mason K, Reno MJ, Blakely L, Vejdan S, Grijalva S. A deep neural network approach for behind-the-meter residential PV size, tilt and azimuth estimation, Solar Energy (2020) 196: 260-269. Massaro A, Starace G. Advanced and Complex Energy Systems Monitoring and Control: A Review on Available Technologies and Their Application Criteria. Sensors (2022), 22:4929. Mateo Romero HF, González Rebollo MÁ, Cardeñoso-Payo V, Alonso Gómez V, Redondo Plaza A, Moyo RT, Hernández-Callejo L. Applications of Artificial Intelligence to Photovoltaic Systems: A Review. Applied Sciences (2022) 19: 10056. Matsuo Y, LeCun Y, Sahani M, Precup D, Silver D, Sugiyama M, Uchibe E, Morimoto J. Deep learning, reinforcement learning, and world models. Neural Networks (2022) 152: 267-275. Mellit A, Kalogirou SA. Machine learning and deep learning for photovoltaic applications. Artificial Intelligence for Smart Photovoltaic Technologies (2022): 1. Miranda E, Fierro JF, Narváez G, Giraldo LF, Bressan M., Prediction of site-specific solar diffuse horizontal irradiance from two input variables in Colombia. Heliyon (2021) 7: 08602. Mittal S. A Review on forecasting the photovoltaic power Using Machine Learning. Journal of Physics: Conference Series (2022) 2286: 012010. Mustafa Z, Awad AS, Azzouz M, Azab A. Fault identification for photovoltaic systems using a multi-output deep learning approach. Expert Systems with Applications (2022) 211:118551. Narvaez G, Giraldo LF, Bressan M, Pantoja A. Machine Learning for Site-Adaptation and Solar Radiation. Forecasting (2021) 167: 333-342.

58

Ekaterina A. Engel

Natsheh E, Samara S. Tree search fuzzy NARX neural network fault detection technique for PV systems with IoT support. Electronics (2020) 9:1087. Nespoli A, Ogliari E, Leva S, Massi Pavan A, Mellit A, Lughi V, Dolara A. Day-Ahead Photovoltaic Forecasting: A Comparison of the Most Effective Techniques. Energies (2019) 12: 1621. Nguyen G, Dlugolinsky S, Bobák M, Tran V, López García Á, Heredia I, Malík P, Hluchý L. Machine Learning and Deep Learning frameworks and libraries for large-scale data mining: a survey. Artif Intell Rev (2019) 52:77–124. Nguyen-Duc T, Le-Viet T, Nguyen-Dang D, Dao-Quang T, Bui-Quang M. Photovoltaic Array Reconfiguration under Partial Shading Conditions Based on Short-Circuit Current Estimated by Convolutional Neural Network. Energies (2022) 15: 6341. Notton G, Voyant C, Fouilloy A, Duchaud JL, Nivet ML. Some Applications of ANN to Solar Radiation Estimation and Forecasting for Energy Applications. Appl. Sci. (2019) 9: 209. Official reinforcement learning environment for demand response and load shaping. Available online: https://github.com/intelligent-environments-lab/CityLearn (05.10.2022). Omar M, Dolara A, Magistrati G, Mussetta M, Ogliari E, Viola F. Day-ahead forecasting for photovoltaic power using artificial neural networks ensembles. In Proceedings of IEEE International Conference on Renewable Energy Research and Applications (ICRERA), Birmingham, (2016) 20-23. Open Neural Network Exchange. Available online: https://onnx.ai/ (11.09.2022). OpenAI. Gym A toolkit for developing and comparing reinforcement learning algorithms. Available online: https://gym.openai.com (13.09.2022) Osmani K, Haddad A, Jaber H, Lemenand T, Castanier B, Ramadan M, Mitigating the effects of partial shading on PV system’s performance through PV array reconfiguration: A review. Thermal Science and Engineering Progress (2022) 31: 101280. Pahwa K, Sharma M, Saggu MS, Mandpura AK. Performance evaluation of machine learning techniques for fault detection and classification in PV array systems. In 7th International Conference on Signal Processing and Integrated Networks (SPIN) (2020) 791-796. Phan BC, Lai YC, Lin CE. A Deep Reinforcement Learning-Based MPPT Control for PV Systems under Partial Shading Condition. Sensors (2020) 20: 3039. Pierdicca R, Paolanti M, Felicetti A, Piccinini F, Zingaretti P. Automatic Faults Detection of Photovoltaic Farms: solAIr, a Deep Learning-Based System for Thermal Images. Energies (2020) 13: 6496. Pierdicca R, Paolanti M, Felicetti A, Piccinini F, Zingaretti P. Photovoltaic Thermal Images Dataset. Available online: https://vrai.dii.univpm.it/content/photovoltaic-thermalimages-dataset (12.09.2022).

Advances in the Artificial Intelligence Models for Photovoltaic Systems

59

Pillai DS, Rajasekar N. Metaheuristic algorithms for PV parameter identification: A comprehensive review with an application to threshold setting for fault detection in PV systems. Renewable and Sustainable Energy Reviews (2018) 82:3503-3525. Pombo DV, Bacher P, Ziras C, Bindner HW, Spataru SV, Sørensen PE. Benchmarking physics-informed machine learning-based short term PV-power forecasting tools. Energy Reports (2022) 8: 6512-6520. Pombo DV, Bindner HW, Spataru SV, Sørensen PE, Bacher P. Increasing the Accuracy of Hourly Multi-Output Solar Power Forecast with Physics-Informed Machine Learning. Sensors (2022) 22: 749. Pymgrid library. Available online: https://github.com/Total-RD/pymgrid (15.10.2022) Ram JP, Babu TS, Dragicevic T, Rajasekar N. A new hybrid bee pollinator flower pollination algorithm for solar parameter estimation. Energy Convers Manag (2017) 135: 463-476. Ren X, Zhang F, Zhu H, Liu Y. Quad-kernel deep convolutional neural network for intrahour photovoltaic power forecasting. Applied Energy (2022) 323: 119682. Sairam S, Seshadhri S, Marafioti G, Srinivasan S, Mathisen G, Bekiroglu K. Edge-based Explainable Fault Detection Systems for photovoltaic panels on edge nodes. Renewable Energy (2022) 185:1425-1440. Sama A. Hello tomorrow — I am a hybrid quantum machine learning, medium (2021). Available Online: https://andisama.medium.com/hello-tomorrow-i-am-a-hybrid-qmlb70751e36142. (22.09.2022). Sammut C, Webb GI. Encyclopedia of machine learning and data mining. Springer (2017) 393-402. Scilab 6.1.1 Available online: https://www.scilab.org/download/scilab-6.1.1 (15.10.2022) Sengupta M, Xie Y, Lopez A, Habte A, Maclaurin G, Shelby J. The national solar radiation data base (NSRDB). Renewable and Sustainable Energy Reviews (2018) 89: 51-60. Shaheen AM, El-Seheimy RA, Xiong G, Elattar E, Ginidi AR. Parameter identification of solar photovoltaic cell and module models via supply demand optimizer, Ain Shams Engineering Journal (2022) 13: 101705. Shahhosseini M, Hu G, Pham H. Optimizing ensemble weights and hyperparameters of machine learning models for regression problems. Machine Learning with Applications (2022)7: 100251. Sizkouhi AM, Aghaei M, Esmailifar SM. A deep convolutional encoder-decoder architecture for autonomous fault detection of PV plants using multi-copters. Sol. Energy (2021) 223: 217–228. Solar energy environmental mapper (solar mapper) web-based GIS application. Available online: https://solarmapper.anl.gov/ (12.09.2022). Solis-Cisneros HI, Sevilla-Camacho PY, Robles-Ocampo JB, Zuñiga-Reyes MA, Rodríguez-Resendíz J, Muñiz-Soria J, Hernández-Gutiérrez CA. A dynamic reconfiguration method based on neuro-fuzzy control algorithm for partially shaded PV arrays. Sustainable Energy Technologies and Assessments (2022) 52:102147.

60

Ekaterina A. Engel

Starzyński J, Zawadzki P, Harańczyk D. Machine Learning in Solar Plants Inspection Automation. Energies (2022) 15:5966. Suresh V, Janik P, Rezmer J, Leonowicz Z. Forecasting Solar PV Output Using Convolutional Neural Networks with a Sliding Window Algorithm. Energies (2020) 13: 723. Terrén-Serrano G, Bashir A, Estrada T, Martínez-Ramón M. Girasol, a sky imaging and global solar irradiance dataset. Data in Brief (2021) 35:106914. The module of the modified fuzzy neural net. M.: Federal Service for Intellectual Property (Rospatent), Certificate about State registration of software No. 2021681065, 2021. The

SOLETE dataset. Available online: https://data.dtu.dk/articles/dataset/The_SOLETE_dataset/17040767(12.09.2022).

Tina GM, Ventura C, Ferlito S, De Vito S. A state-of-art-review on machine-learning based methods for PV. Applied Sciences (2021) 16: 7550. Tovar M, Robles M, Rashid F. PV Power Prediction, Using CNN-LSTM Hybrid Neural Network Model. Case of Study: Temixco-Morelos, México. Energies (2020) 13: 6512. Uehara G, Rao S, Dobson M, Tepedelenlioglu C, Spanias A. Quantum Neural Network Parameter Estimation for Photovoltaic Fault Detection. In 12th International Conference on Information, Intelligence, Systems & Application. IISA (2021): 1–7. Vandemark A. Real-time series resistance monitoring of solar PV modules by communicating limited remote measurements to cloud-based algorithm via API. United States (2020). Venkatesh SN, Sugumaran V. A combined approach of convolutional neural networks and machine learning for visual fault classification in photovoltaic modules. Proc. Inst. Mech. Eng. Part O J. Risk Reliab. (2022) 236: 148–159. Wang MH, Lin ZH, Lu SD. A Fault Detection Method Based on CNN and Symmetrized Dot Pattern for PV Modules. Energies (2022) 15:6449. Wang X, Lin X, Dang X. Supervised learning in spiking neural networks: A review of algorithms and evaluations. Neural Networks (2020) 125: 258-280. Wang Y, Zou H, Chen X, Zhang F, Chen J. Adaptive Solar Power Forecasting based on Machine Learning Methods. Appl. Sci. (2018) 8: 2224. Warden P, Situnayake D. TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-low-power Microcontrollers. O’Reilly Media, Inc. (2021). Wu YK, Huang CL, Phan QT, Li YY. Completed Review of Various Solar Power Forecasting Techniques Considering Different Viewpoints. Energies (2022) 15: 3320. Xu L, Bai L, Bao H, Jiang J. Parameter Identification of Solar Cell Model Based on Improved Artificial Bee Colony Algorithm. 13th International Conference on Advanced Computational Intelligence (ICACI) (2021) 239-244. Youssef A, El-Telbany M, Zekry A. The role of artificial intelligence in photo-voltaic systems design and control: A review. Renewable and Sustainable Energy Reviews (2017) 78: 72-79.

Advances in the Artificial Intelligence Models for Photovoltaic Systems

61

Yu K, Qu B, Yue C, Ge S, Chen X, Liang J. A performance guided JAYA algorithm for parameters identification of photovoltaic cell and module. Applied Energy (2019) 237:241–257. Yu X, Wu X, Luo W. Parameter Identification of Photovoltaic Models by Hybrid Adaptive JAYA Algorithm. Mathematics (2022) 10: 183. Zefri Y, Sebari I, Hajji H, Aniba G. Developing a deep learning-based layer-3 solution for thermal infrared large-scale photovoltaic module inspection from orthorectified big UAV imagery data. Int. J. Appl. Earth Obs. Geoinf. (2022) 106: 102652. Zellagui M, Settoul S, El-Bayeh CZ, Belbachir N. Chaos Game Optimization Algorithm for Parameters Identification of Different Models of Photovoltaic Solar Cell and Module, J. Ren. Energies (2022) 1: 245. Zhang H, Zhu T. Stacking Model for Photovoltaic-Power-Generation Prediction. Sustainability (2022) 14: 5669 Zhang J, Verschae R, Nobuhara S, Lalonde JF. Deep photovoltaic nowcasting. Sol. Energy (2018) 176: 267–276. Zhen Z, Liu J, Zhang Z, Wang F, Chai H, Yu Y, Lu X, Wang T, Lin Y. Deep learning based surface irradiance mapping model for solar PV power forecasting using sky image. IEEE Trans. Ind. Appl. (2020) 56: 3385–3396. Zhou Y, Zheng S, Zhang G. Machine learning-based optimal design of a phase change material integrated renewable system with on-site PV, radiative cooling and hybrid ventilations—study of modeling and application in five climatic regions. Energy (2020) 192: 116608. Zjavka L, Snášel V. PV Energy Prediction in 24 h Horizon Using Modular Models Based on Polynomial Conversion of the L-Transform PDE Derivatives in Node-by-NodeEvolved Binary-Tree Networks. Eng. Proc. (2022) 18: 34. Zjavka L. PV power intra-day predictions using PDE models of polynomial networks based on operational calculus. IET Renew. Power Gener. (2020) 14: 1405–1412.

Chapter 2

Recent Advances in Photovoltaic Materials and Technology Sukla Basu∗, PhD (Tech.)

Department of Electronics & Communication Engineering, Kalyani Government Engineering College, Kalyani, West Bengal, India

Abstract Energy demand of the globe is ever increasing with the progress of human civilization. Conventional energy resources, such as fossil fuels will be exhausted in the not-too-distant future. Besides, the production of energy from fossil fuels accounts for the upward trend in carbon dioxide and other greenhouse gas emission leading to climate change and global warming. Solar energy has the potential to meet the energy demand of globe due to its abundance provided this energy can be harnessed efficiently. Photovoltaic is an important clean energy production technology that can convert solar light to electrical power directly without burning non-renewable carbon fuels like coal, oil and natural gas. The photovoltaic cell or solar cell was first developed in 1954 using a diffused Si (Silicon) p-n junction. Silicon based solar cells were the firstgeneration solar cells grown on Si wafers, mainly single crystals. Second generation solar cells are mostly thin film solar cells made of amorphous Si, or compound semiconductors like CdTe (Cadmium Telluride) and CIGS (Copper Indium Gallium di-Selenide). Silicon is mostly used material for fabrication of solar cells still now mainly due to its abundance on earth crust and highly developed Si technology and processes. Limited efficiency of the solar power to electrical power conversion in solar cells is a major disadvantage. Considerable research ∗

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

64

Sukla Basu works have been done over the last few decades for overcoming this disadvantage and improving the performance characteristics of solar cells. Other than Si, a number of inorganic compound semiconductors, perovskites, organic polymers, hybrid materials and multi junctions have been investigated for fabrication of efficient solar cells. Semiconducting organic macromolecules, inorganic nano particles or hybrids are of current research interest for fabrication of solar cells. In this chapter a comprehensive review of research works on solar cell materials and technology in last few decades and comparison of the performance characteristics of various types of solar cells is presented. Main focus is given on the advancement in this field over last ten years.

Keywords: Photovoltaic cells, Photovoltaic Technology

Solar

cells,

Photovoltaic

materials,

Introduction Solar energy is regarded as the most promising source of clean energy for the coming days. This natural source of energy can be converted to electrical energy with a simple mechanism. However, mainly due to low conversion efficiency of the solar cells, use of solar power is limited at present. American researchers—Gerald Pearson, Daryl Chapin, and Calvin Fuller— demonstrated a Si solar cell capable of a 6 percent energy-conversion efficiency when used in direct sunlight in 1954. Considerable research efforts have been made since then to overcome the constraint of low efficiency. By the late 1980s Si cells, as well as cells made of gallium arsenide, with efficiencies of more than 20 percent had been fabricated. With additional accessories like lenses sunlight can be concentrated onto the solar cell surface to achieve an efficiency of about 37 percent. By connecting cells of different semiconductors optically and electrically in series, even higher efficiencies are possible, but at increased cost and added complexity. In general, solar cells of widely varying efficiencies and costs are now available (“Solar Cell - Solar Panel Design”, 2019). Also, a number of different materials are being used for fabrication of these solar cells. Depending on the materials and fabrication technologies used, these cells can be classified into four generations. In the following sections, important candidates of these four generations are briefly described after giving a short account of the basic operating principle of a solar cell.

Recent Advances in Photovoltaic Materials and Technology

65

Basics of Solar Cells The basic steps involved in the conversion of solar energy into electrical energy are as follows (Figure 1): • Absorption of solar energy by the cell • Generation of charge carriers (electron and holes) due to optical energy transfer • Separation of charge carriers • Subsequent collection of the carriers in the respective electrodes and thereby production of potential difference between the electrodes • Supply of current to the load connected across the electrodes.

Figure 1. Schematic representation of a Si p-n junction solar cell (Sze and Lee, 2013).

Efficiency of a solar cell is defined as the ratio of energy output from the solar cell to input energy from the sun. The efficiency depends primarily on the spectrum of the incident sunlight and band gap of the material of the solar cell. Besides, reflectivity of the cell surface, and temperature of the cell are other important factors those affect the efficiency of a cell. Therefore, conditions under which efficiency is measured must be carefully controlled to compare the performance of one device to another (“Solar Cell Efficiency PV Education” 2019).

66

Sukla Basu

First Generation Solar Cells First generation solar cells are produced on either mono crystalline Si or polycrystalline Si wafers. Mono crystalline Si is produced by a process called Czochralski process (Würfel et al., 2016). The large single crystal wafer production requires precise processing steps. The efficiency of mono crystalline Si solar cells (Figure 2) lies between 17% - 18% (Sharma et al., 2015). High cost and the sophisticated technological steps have led to use polycrystalline Si instead of the single crystal Si wafers. The polycrystalline Si is produced by cooling molten Si in a graphite mould and fabrication steps are relatively simple and less expensive than that of mono crystalline Si wafer. However, polycrystalline Si solar cells are less efficient ~12% - 14% than mono crystalline solar cells (Sharma et al., 2015). One of the important advantages of first- generation solar cells is the easy availability of raw material, Si. Other advantages are highly developed Si processing technology, well understood Si device physics and long- term stability of the Si solar cells. Main disadvantage of first- generation solar cell is low solar power absorption coefficient of Si since it is an indirect bandgap semiconductor.

Figure 2. An energy band diagram showing the major energy leak pathways: (1) recombination loss; (2, 3 and 4) thermal loss; (5) photons with insufficient energy cause non-absorption (Almomani et al., 2022).

Recent Advances in Photovoltaic Materials and Technology

67

Second Generation Solar Cells Second Generation solar cells are thin film solar cells. Silicon wafers of firstgeneration solar cells have to be at least 50 µm thick for effective solar power absorption, when no optical enhancement techniques are used to improve the effective absorption. Thin-film solar cells have very thin light absorbing layers, generally of the order of 1 µm thickness (Chopra, et al., 2004) which is deposited on a substrate. Amorphous (a-Si) thin film solar cells can be manufactured at a low processing temperature, thereby permitting the use of various low cost, polymer and other flexible substrates. Amorphous Si (a-Si) has a direct band gap and uses only 1% of the material (Si) needed for crystalline Si cells (Ong et al., 2018). Therefore, a-Si solar cells are comparatively cheaper than first generation solar cells. The major matter of concern for a-Si solar cell is its poor and almost unstable efficiency. Besides amorphous Si (a-Si), two-component (binary) materials like GaAs, InP, CdTe etc. are attractive for thin-film solar cells. Of these, GaAs, InP and their derived alloys and compounds are ideal for photovoltaic applications, but they are too expensive to be used for large-scale commercial applications. CdTe thin film solar cells are economically viable. They have high optical absorption coefficient and chemical stability; moreover, they can be made on polymer substrates and thus flexible. However, a limited resource of cadmium and environmental hazards associated with its use are the main issues with this CdTe technology (Sharma et al., 2015). Among quaternary compound semiconductors, CIGS, comprising of the four elements, namely: Copper, Indium, Gallium and Selenium is another potential candidate for thin film solar cell. Compared to the CdTe thin film solar cell, CIGS cells have higher efficiency. Different types of substrate materials can be used in CIGS solar cells. The substrate materials those can be used in CIGS cells include glass, polymers and metals like steel, aluminum and molybdenum. One of the advantages of CIGS-based solar cells is that the band gap can be engineered by adjusting the Ga/In ratio to match the solar spectrum. CIGS absorber layer can absorb most parts of the solar spectrum with a thickness of 1 μm (Ong et al., 2018). Hence, total thickness of the solar cell is only around 2.5 μm. Thin layer device means reduction in raw material usage and thereby low production cost. Another advantage of CIGS thin film solar cells is its prolonged life without considerable degradation.

68

Sukla Basu

Among thin film solar cells, efficiencies of amorphous Si(a-Si), CdTe, and CIGS cell are 14%, 22.1%, and 22.6%, respectively, as reported by the researchers (Ong et al., 2018) .

Third Generation Solar Cells A number of new technologies have been developed over recent years to improve the performance characteristics of solar cells. Inorganic materials other than Si (Badawy, 2015; Ebhota and Jen, 2018; Ahmad et al., 2022; Shi et al., 2015) are used in third generation solar cells. Besides, various organic materials, hybrid materials those are composites of both organic and inorganic materials are also used in the solar cells of third generation. All of these technologies aim at developing low-cost, high efficiency solar cells.

Polymer Solar Cells There are two main types of polymer solar cells: polymer: fullerene solar cells and hybrid polymer solar cells (Yan and R. Saunders, 2014). Polymer: fullerene solar cells have photoactive layers comprising of a semiconducting polymer and a fullerene which is an allotrope of Carbon. Semiconducting polymers are the electron donors and fullerenes are good electron acceptors. Around 12% efficiency is reported for this type of solar cells. In the case of hybrid polymer solar cells, the fullerene is replaced by a semiconducting inorganic nanoparticle. The best hybrid polymer solar cell in terms of power conversion efficiency is PDTBT: PbS0.4Se0.6 solar cell which achieves 5.5% efficiency as reported in literature (Yan et al., 2014). Polymer solar cells are flexible, lightweight and durable. They are 50% less expensive than conventional Si solar cells (Sharma et al., 2015). However, they cannot be used in high temperature conditions. Perovskite Based Solar Cells Perovskites are a class of compounds defined by the formula ABX3 where X represents either O, C, N or a halogen such as I, Br, Cl. A and B are cations of different sizes (Sharma, Jain, and Sharma 2015). The most common perovskites currently used for solar cell applications are CH3NH3PbI3, CH3NH3PbBr3 and the mixed halide system, CH3 NH3PbI2X perovskites (X=Cl, Br, I). Conventional Si based solar cells need expensive, multiple processing steps and require high temperatures (>1000˚C) and vacuums

Recent Advances in Photovoltaic Materials and Technology

69

facilities, whereas perovskite cells are of low cost due to the low temperature solution methods of processing and the absence of rare elements. The high efficiency of lead-based perovskite solar cells has outperformed the more established and expensive solar cell technology of silicon, gallium arsenide and cadmium telluride. The field of perovskite solar cells has matured rapidly, and the efficiency is now reaching more than 25% (Ahmad et al., 2022). The rapid maturity of this field is contributed by superior optoelectronic properties of perovskites which include excellent light absorption coefficient, long carrier lifetime, high charge carrier mobility, and low defect density.

Dye-sensitized Solar Cells (DSSC) DSSCs are based on a porous, thin film of a wide-bandgap semiconductor oxide modified by dye molecules. The sponge-like characteristics and increased surface area of this type of film enhances the light absorption. A typical DSSC consists of transparent conductive oxide (TCO), semiconductor oxide, dye sensitizer, electrolyte and counter electrode. Soda lime glass coated with fluorine tin oxide or indium tin oxide is typically used as TCO. The central part of a DSSC device consists of a thick nanoparticle film made of semiconductor oxide materials like TiO2 that provides a large surface area. A single layer of dye molecules acts as a light absorber and is interspersed between TiO2 particles. The dye sensitizers used in DSSCs include metal complex sensitizers, metal-free organic sensitizers and natural sensitizers. An ideal dye sensitizer should adsorb a wide range of wavelengths and possess high thermal stability. Liquid electrolytes are used to regenerate the dye after electron injection into the conduction band of the semiconductor oxide layer and also act as a charge transport medium for the transfer of positive charge toward the counter electrode. The counter electrode transports the electron that arrives from the external circuit back to the electrolyte system and is typically made of Platinum. The overall conversion efficiency of DSSCs (Figure 3) is proportional to the injection of electrons in wide-bandgap nanostructured semiconductors. The certified efficiency record is approximately 11.1% (Suhaimi et al. 2015). The manufacturing cost of DSSCs is approximately 1/3 to 1/5 times that of silicon solar cells (Sharma et al., 2015). However, there are certain challenges like degradation of liquid electrolyte and degradation of dye molecules after exposure to ultraviolet and infrared radiation leading to a decrease in the lifetime and stability of the cells.

70

Sukla Basu

Figure 3. Dye-sensitized Solar Cells (DSSC) (Kosyachenko, 2015)

Multi-junction Solar Cells Single-junction solar cells are typically made of the same semiconductor material, while multi-junction solar cells commonly use three separate semiconductors of different band gap energy values . These layers are capable of absorbing different wavelengths of incoming sunlight. Thus they are more efficient for converting sunlight into electricity than single-junction cells. Band gap of the materials from the top to the bottom going to be smaller and smaller. It allows to absorb and convert the photons that have energies greater than the band gap of that layer and less than the band gap of the higher layer. Typically gallium indium phosphide (GaInP), indium gallium arsenide (InGaAs), and germanium (Ge) are used (Yamaguchi et al., 2021). The fundamental limitation of multi-junction solar cells is availability of materials with optimal band gaps and low defect densities. Quantum Dot [QD] Solar Cells Quantum dot solar cells are composed of semiconducting particle of size few nanometers, these are called quantum dots. In a QD solar cell, quantum dots are the key absorbing photovoltaic material. Energy band gaps of quantum dots can be changed by changing their size only without changing the underlying material or fabrication method. In typical wet chemical synthesis, the time duration and temperature of synthesis determine the dot size. On the contrary, the band gap is fixed for a given bulk material. This flexibility of

Recent Advances in Photovoltaic Materials and Technology

71

quantum dots makes them really attractive for fabrication of solar cells which can absorb solar energy over a broad spectrum of wavelengths. With the advancement of nanotechnology, the design of QD solar cells aims at replacing the bulk materials those are presently used for solar cells. Many choices and combinations of QD materials have been reported in literature. The QD solar cells can be made of any type of semiconducting material as long as its absorption is compatible with the solar spectrum. In recent years lead chalcogenide QDs, lead halide perovskite QDs, and lead-free QDs (Albaladejo‐Siguan et al., 2021) are of research interest. Chalocgenide QDs made from lead sulfide, selenide, and telluride (PbX: PbS, PbSe, and PbTe) have properties that make them particularly useful as solar cells. Of the three lead chalcogenides, lead sulfide (PbS) is the most studied material in photovoltaic applications. The bandgap of bulk PbS is 0.4 eV, which can be increased up to 2.3 eV by reducing the QD diameter to 1.8 nm. As reported in recent literature, power conversion efficiency of PbS QD is 14% in 2020 (Albaladejo-Siguan et al., 2021). Solar cells based on perovskite QDs are relatively new, CsPbI3 QD devices with power conversion efficiency of 13.43% (Swarnkar et al., 2016) and 15.1% (Hu et al., 2021) are recently reported in literature. In 2019, Hao et al. reported QDSCs using hybrid Caesiumformamidinium (FA) lead iodide [Cs0.5FA0.5PbI3]. These QDs reach an efficiency of 16.6% (Hao et al., 2020). Other perovskite compositions have also been explored for fabrication of efficient solar cells. The highest reported conversion efficiency of solar cells made from leadfree nanocrystals such as binary compounds like CdSe and InP, and tertiary compound like AgBiS2 (Öberg et al., 2020) are currently around 13%. The main disadvantage of QD solar cells is ambient factors such as oxygen, water, high temperature or high intensity of illumination can decrease the lifetime of QD solar cells. Beyond these external factors, interaction of QDs with other materials present in the photovoltaic devices can cause degradation of their performance. Intensive research works for overcoming these disadvantages of QD solar cells are going on.

Fourth Generation Solar Cells The solar cells of this generation are hybrid type. Thin polymer layers and metal nanoparticles, as well as various metal oxides, carbon nanotubes, graphene, and their derivatives are used in fourth generation solar cells (Luque and Hegedus, 2011). Among all these materials graphene is the most important nanomaterial due to its unique properties such as high carrier mobility, low resistivity, transmittance and 2D lattice packing.

72

Sukla Basu

Graphene-based materials are established as efficient candidates to replace or modify the existing components in solar cells (Adil et al., 2018). These materials can be used for a number of purposes such as transparent conductive electrodes, semiconducting layers, sensitizers, electrolytes, electron-hole transport layers, and counter electrodes in solar cells. So far, the various graphene-based materials that have been investigated have shown enhanced or comparable performances to those of conventional devices. Since the properties of graphene are very much dependent on its fabrication process, a judicious choice of methods is essential for a particular application. Graphene based solar cells are also protected from environmental degradation due to the packed 2D lattice structure of graphene. Graphene's major disadvantage is its hydrophobic nature which negatively affects the processing of the graphene-based device in solution. As reported in recent literature the energy conversion efficiency can exceed 20% for graphene-based perovskite solar cells (Gong et al., 2021).

Conclusions A number of different generation solar cells are discussed in the present chapter. Table 1 gives a comparison of four generations of solar cells presented here. Still now Si is the most used material for solar cells due to its easy availability and mature processing technology. Considering advantages and disadvantages of various types of solar cells in respect of solar power conversion efficiency, availability of material, cost of production, durability, environmental effects, it is seen that different types of perovskite solar cells are very promising candidates for the production of inexpensive, high efficiency solar cells in the coming years. Table 1. Comparison of solar cells of different generations Type

Materials used

1st generation solar cells

Mono Crystalline /Poly crystalline Si

2nd generation solar cells

Thin Film Si/Compound semiconductor

Advantage Easy availability of Si, highly developed Si processing technology. Low raw material usage and thereby low production cost. Long life of CIGS solar cells.

Disadvantage

Maximum Efficiency

Low power absorption coefficient of Si.

~ 18% [Mono crystalline solar cell]

Poor and unstable efficiency of aSi solar cells. Toxic nature of CdTe cells.

~22% (CdTe/CIGS cell)

Recent Advances in Photovoltaic Materials and Technology Type

Materials used

3rd generation solar cells

Organic/Inorganic semiconductor/Hybrid

4th generation solar cells

Hybrid in nature. Thin polymer layers and metal nanoparticles, as well as various metal oxides, carbon nanotubes, graphene, and their derivatives.

Advantage Generally low processing temperature and low cost. Possibility of absorption of wide spectrum of solar energy by quantum dot of the same compound material. Enhanced or comparable performances to those of conventional devices. Low environmental degradation of graphene-based cells.

Disadvantage

Maximum Efficiency

Degradation of performance by environmental factors.

~ 25% (Perovskite solar cell)

Optimization of defect densities of graphene for efficient performance is difficult.

~ 20%

73

Disclaimer None

References Adil SF, Khan M, Kalpana D. Graphene-based nanomaterials for solar cells. In Multifunctional Photocatalytic Materials for Energy (2018): 127-152. Ahmad S, Kazim S, Grätzel M eds. Perovskite Solar Cells: Materials, Processes, and Devices (2022) John Wiley & Sons. Albaladejo‐Siguan M, Baird EC, Becker‐Koch D, Li Y, Rogach AL, Vaynzof Y. Stability of quantum dot solar cells: A matter of (life) time. Advanced Energy Materials, (2021) 11(12): 2003457. Almomani MS, Ahmed NM, Rashid M, Ibnaouf KH, Aldaghri OA, Madkhali N, Cabrera H. Performance improvement of graded bandgap solar cell via optimization of energy levels alignment in Si quantum dot, TiO2 nanoparticles, and porous Si. Photonics (2022) 9(11): 843. Asim N, Sopian K, Ahmadi S, Saeedfar K, Alghoul MA, Saadatian O, Zaidi SH. A review on the role of materials science in solar cells. Renewable and sustainable energy reviews (2012) 16(8): 5834-5847. Chopra KL, Paulson PD, Dutta V. Thin-film solar cells: an overview. Prog. Photovoltaics Res. Appl (2004) 12(23): 69-92.

74

Sukla Basu

Ebhota WS, Jen TC. Efficient low-cost materials for solar energy applications: roles of nanotechnology. Recent developments in photovoltaic materials and devices (2018). Gong K, Hu J, Cui N, Xue Y, Li L, Long G, Lin S. The roles of graphene and its derivatives in perovskite solar cells: A review. Materials & Design, (2021) 211: 110170. Hao M, Bai Y, Zeiske S, Ren L, Liu J, Yuan Y, Zarrabi N, Cheng N, Ghasemi M, Chen P, Lyu M. Ligand-assisted cation-exchange engineering for high-efficiency colloidal Cs1− x FA x PbI3 quantum dot solar cells with reduced phase segregation. Nature Energy (2020) 5(1): 79-88. Hu L, Zhao Q, Huang S, Zheng J, Guan X, Patterson R, Kim J, Shi L, Lin CH, Lei Q, Chu D. Flexible and efficient perovskite quantum dot solar cells via hybrid interfacial architecture. Nature communications (2021) 12(1): 466. Kosyachenko LA Ed. Solar Cells: Dye-Sensitized Devices. BoD–Books on Demand (2011). Luque A, Hegedus S. (Eds.). Handbook of photovoltaic science and engineering. John Wiley & Sons (2011). Oberg VA, Johansson MB, Zhang X, Johansson EM. Cubic AgBiS2 colloidal nanocrystals for solar cells. ACS Applied Nano Materials (2020) 3(5): 4014-4024. Ong KH, Agileswari R, Maniscalco B, Arnou P, Kumar CC, Bowers JW, Marsadek MB. Review on substrate and molybdenum back contact in CIGS thin film solar cell. International Journal of Photoenergy (2018). Sharma S, Jain KK, Sharma A. Solar cells: in research and applications—a review. Materials Sciences and Applications (2015) 6(12): 1145. Shi D, Zeng Y, Shen W. Perovskite/c-Si tandem solar cell with inverted nanopyramids: realizing high efficiency by controllable light trapping. Scientific reports (2015) 5(1): 1-10. Solar Cell Efficiency | PVEducation. Pveducation.org. https://www.pveducation.org/pvcdrom/solar-cell-operation/solar-cell-efficiency (2019). Solar Cell Solar Panel Design. In Encyclopædia Britannica. https://www.britannica.com/technology/solar-cell/Solar-panel-design (2019). Suhaimi S, Shahimin MM, Alahmed ZA, Chyský J, Reshak AH. Materials for enhanced dye-sensitized solar cell performance: Electrochemical application. Int. J. Electrochem. Sci (2015) 10(4): 2859-2871. Swarnkar A, Marshall AR, Sanehira EM, Chernomordik BD, Moore DT, Christians JA, Chakrabarti T, Luther JM. Quantum dot–induced phase stabilization of α-CsPbI3 perovskite for high-efficiency photovoltaics. Science (2016) 354(6308):92-95. Sze SM. Semiconductor devices: physics and technology. John Wiley & Sons (2008). Yamaguchi M, Dimroth F, Geisz JF, Ekins-Daukes NJ. Multi-junction solar cells paving the way for super high efficiency. Journal of Applied Physics (2021) 129(24): 240901. Yan J, McNaughter PD, Wang Z, Hodson N, Chen M, Cui Z, O'Brien P, Saunders BR Controlled aggregation of quantum dot dispersions by added amine bilinkers and effects on hybrid polymer film properties. RSC advances (2015) 5(116): 9551295522.

Chapter 3

Carbon Nanodots in Photovoltaic Cells, a Solar Energy Harvester: A Critical Review Biswajit Gayen1,*, PhD Pijush Dutta2, PhD and Chayan Goswami1, MSc 1Department

of Chemistry, Greater Kolkata College of Engineering & Management, Baruipur, West Bengal., India 2Department of ECE, Greater Kolkata College of Engineering & Management, Baruipur, West Bengal., India

Abstract Carbon nanodots, or carbon dots (CDs) have well established themselves with their widespread applications in diverse fields of research and development. More accurately, its effective presence has been shown in the field of photovoltaic devices. Very recently, in the domain of sustainable green energy, specifically for solar energy, a doped carbon dots embedded-FRET based device has bloomed with-outstanding environment-friendly and high efficiency characteristics with very few limitations like photocurrent generation. Carbon nano dots embedded hybrid devices have initiated an impressive revolution in the solar sector from high cost-to-energy ratio to low-cost-energy devices with their magnificent sustainability. Hence, there is a significant opportunity for further research and development. Herein, we endow with a brief overview of current state-of-art carbon nano dots devices with new ideas for further modification in view of efficiency enhancement for carbon

Corresponding Author’s E-mail: [email protected].

∗

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

76

Biswajit Gayen, Pijush Dutta and Chayan Goswami nano dots incorporating solar energy harvesting systems which may be improvised in future.

Keywords: carbon dots, photovoltaic cell, solar cell, FRET, environment friendly

Introduction Demand and supply of energy have an immense importance in our universe from the start of civilization to till date. In the Stone Age, they first invented fire with stones and entered into the domain to control energy sources. From that first day to present, we are working for better, eco-friendly, sustainable, low-cost renewable energy sources and their cent percent utilization. Worldwide, all leading countries with their governments, a number of research institutes, groups & companies working together to achieve the desired goal (Awasthi et al., 2020). Hence, a steady growth in this field is our prior importance in encouraging technological developments and the use of alternative energy sources, which will be accelerated our growth. Solar energy is the most affordable among others as an alternative source of energy. It makes itself unique by its own characteristics, like the most abundant in nature; sunlight is available throughout the year on most of our earth’s surface, no need to pay any penny for this long-term natural resources and is completely eco-friendly (Nayak et al., 2019). Presently, Photo-Voltaic (PV) cell technology is one of the most used methodologies to convert solar energy to electrical energy, direct current in an optimum way. Worldwide, every nation is very much concerned about green sources of energy because all conventional fossil fuel energy sources emit CO2, the main component of greenhouse gases which is caused by global warming and unpredictable climate changes. These uncertain climate changes may lead to danger for all types of living and non-living matters. Whereas solar energy has the power to solve all kinds of energy related problems by its own (Sing et al., 2021). Solar energy has a few limitations, like non-availability of sunlight at night, insufficient sunlight all over the day, non-regular intensity of radiant energy depending on different regions, weathers, and seasons (Ojo et al., 2019). In spite of those demerits, we can overcome the above limitations by proper technological development in the field of storing solar energy for night and further use. Statistical reports have been expressed that 88% of solar energy has returned back to the universe without any impact on the earth’s surface. Out

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

77

of these 12% solar energy, the green plant utilizes a maximum of 11% for photosynthesis, ~0.3% is used for heating the earth’s surface and a very marginal level i.e., ~0.015%, is used fruitfully for the conversion of solar energy to electricity. Though every Nation is favored to use solar energy but about 85% of global energy is generated by means of fossil fuels, whereas just ~3.5% energy is produced by solar PV cells (Pastuszak et al., 2022). Hence, broad opportunities remain open for the R&D section in our near future. In spite of the promising prospects of SCs, there are a number of reports on potential health hazards. All the new technologies adopted in SCs are excellent in terms of reducing carbon footprint but can not avoid the serious health issues for humans. Presently, most of the materials are used in the form of nano materials. Metal/metallic complexes, dyes and compounds exhibited serious toxicities and they are very well known for their carcinogenic nature (Ellina et al., 2020). Very commonly used TiO2 nano material (NM) reported for cellular damage and DNA breakage through oxidative pathways. It is also reported for serious hepatic damage, brain tissue damage which causes injury in brain cell by inflammatory and apoptotic pathways (Ghosh et al., 2021). Similarly, a number of materials like Pb, Cd, Pd, Hg, Si and Ga- compounds used in SCs showed toxicities and health issues. In this point of view, advanced research and development work is necessary towards the use of materials those will not only offer better efficiency of the cell but also be green in nature. A number of group already have started their developmental work in this field to overcome the challenge. Very recent work and their reporting exclaimed that quantum dots like graphene quantum dots (GQDs), carbon dots (CDs) have proved their excellent presence with their fantastic characteristics in view of toxicity and health hazards (Kim et. al., 2022). If we go for report analysis of the utilization of solar energy as electrical energy, it reveals that almost 3000% growth in the last decade from 2010 to 2021 and 22% growth in the year 2020 to 2021. In 2021, total solar PV cell energy is generated by an amount of 1000.9 TWh. It will reach 7413.9 TWh by the year of 2030, considering an annual growth rate of 25%. Our nation stretches its remarkable footprint worldwide by reporting the largest solar power plant, Bhadla solar park of 2245 Mw capacity and the 3rd largest solar power plant, Pavagada solar park of 2050 Mw capacity. Presently, India has achieved around 7% of total electricity using solar PV cells, which is a great work in concern of global economy and carbon footprint (Bojec, 2022; Jaganmohan, 2021). The aim of this review is to present a sketch of present development and research on different types of solar cell to different PV technologies applied

78

Biswajit Gayen, Pijush Dutta and Chayan Goswami

for better and efficient devices and special emphasis on sustainable, ecofriendly solar cell.

Solar Cells The world that we live in is running on the energy sources that are not renewable. The resources that we are using to produce energy are also not endless and eco-friendly. Hence, we should focus on renewable energy and how to use these renewable energies as efficiently as possible. In this journey, SCs play the most important leader compared to other alternative sources of energies. A solar cell simply converts solar energy to electrical energy. More conventionally, a solar cell follows the PV cell mechanism where sunlight is triggered solar cell and it produces electricity. A solar cell works on the principle of a photoelectric effect followed by a p-n junction mechanism. Generally, electron and hole separation take place and a certain voltage is generated across the cell (Figure 1). Depending on different types of mechanisms and materials used for solar cells, these are classified into different categories with varied working principles, efficiencies, ecofriendliness & sustainabilities (Kenu et al., 2020).

Figure 1. A typical presentation of solar cell.

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

79

Different Types of SCs Generally, SCs are categorized in three main categories and subdivided as below (Figure 2).

Figure 2. Flow chart for different categorized SCs.

First Generation SCs First generation solar cells or silicon wafer based solar cells introduce two kinds of PV cells. First type is single crystalline solar cells and 2nd type is polycrystalline solar cell.

Monocrystalline Silicon SCs First generation solar cells are made of pure and crystalline silicon materials. The market was dominated by the monocrystalline SCs; more than 80% of the solar cells are of this type. In 2001, Green et al., reported a high efficiency of 25% by crystalline SCs (Green et al., 2001). Due to the use of pure silicon crystals the market price of these crystals based solar cell cost was very high. Hence it is a costly alternative technology to generate energy from sunlight. The Czochralski method was employed for the fabrication of these kinds of cells. Efficiency of these cells is also a function of temperature. The temperature dependency of the efficiency of these solar PV cells at constant intensity of light was very well studied. A maximum of 31% thermodynamic efficiency for the conversion of the photo energy to electrical energy is still to be achieved, though at optimum condition these cells show good efficiency.

80

Biswajit Gayen, Pijush Dutta and Chayan Goswami

Polycrystalline Silicon SCs To reduce the cost of crystalline SCs, people started the use of polycrystalline material in SCs. Zhao et al., (1998) was introduced with polycrystalline cells with honeycomb like structure and reported a poor efficiency, 19.8% compared to crystalline solar cell. Extensive work has been reported because of their poor efficiency. The polycrystalline solar PV cells available in the market are expected to have a maximum lifetime of 27 years. Although polycrystalline solar cells are less expensive than monocrystalline ones, but they do have some serious drawbacks. These cells do not work well in low light, along with the fact that they require a wide installation area, and they are not much thermally stable compared to former. Second Generation SCs Second generation solar cells are mostly based on thin-film technology. The manufacturing cost of the crystalline silicon made first generation solar cells too expensive. To reduce the manufacturing cost the idea was to reduce the amount of silicon. As a result, instead of crystalline silicon, development of solar cells based on thin-films is becoming more familiar and well accepted.

Amorphous Silicon SCs The markets for amorphous silicon (a-Si) based solar cells have been running for more than one decade. These solar cells are very user-friendly and in recent days it has been used in recharging batteries of calculators and other electronic gadgets. In a-Si solar cell, a band gap of 1.7eV is observed due to its amorphous structure offering towards higher efficiency, is around 13.8%. These solar cells can be manufactured with low material consumption and processed at low temperature and most importantly their cost-effectiveness. Copper Zinc Tin Sulphide Solar Cells (CZTS solar cells) The frequent availability and abundance of the metals like Cu, Zn and Sn encourage the modern generation workers for the synthesis of the CZTS and CZTSS (Copper zinc tin sulphide/selenide solar cells). Power conversion efficiency (PCE) of these chalcogenides based solar cells reached up to 11.3% (Green et al., 2019). The major problem of these types of cell is recombination of charges and if the probability of the recombination is controlled, then the efficiency of the cells can be increased. Solubility of CZTS material on

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

81

commonly used solvents is not up to the mark, hence making of a thin film was always remaining a crucial problem. Hence, advanced fabrication method to prepare CZTS based solar cells is essential for their sustainability.

Cadmium Sulphide and Cadmium Telluride Thin Film Technology Solar cells made of CdS and CdTe are well-known due to their high stability. After Silicon, CdTe based solar cells are considered to be the most widely used solar technology. The maximum efficiency of these solar cells can reach up to 21% theoretically and NREL reports the efficiency of 16.5% in the laboratory (Wu et al., 2004). Such high efficiency is a result of perfect band gap energy of 1.45 eV. The excellent efficiency of these types of SCs are attributed based on photon energies that are getting from sunlight, which are very much appropriate for the excitation of the photo-electrons in CdTe materials. Along with the abundance of the constituent material, these SCs are well accepted because of their efficiency, wide range applications and many more, but toxicity of heavy material becomes an environmental issue. Copper Indium Gallium Di-selenide (CIGS) SCs These cells are used for advanced research purposes. For commercial purposes the use of these cells are restricted due to rare availability of the metals like Ga and In. The efficiency of these cells is increased with depositing them with heavy alkali metals like Cs. Efficiency for these cells can exceed 19.9% (Repins et al., 2008). Most of the currently available solar PV cells are fabricated on rigid surface, but in CIGS SCs the fabrications are done on flexible surface like polyamide film, soda lime glass and many others. These flexibilities provide CIGS SCs an extra importance. Third Generation SCs First and second generation solar cells are not useful due to the presence of only one junction, short range of absorption, non-eco-friendly and expensive nature. Recently discovered photo voltaic cells are more than 30% efficient. Most of them are multi junctions and absorb over a wider range of frequencies.

Dye sensitized SCs (DSSC) To avoid the high price of the first generation solar cell and manufacturing difficulties of the second generation solar cells, a relatively low cost, highly

82

Biswajit Gayen, Pijush Dutta and Chayan Goswami

efficient and economically affordable PV cell is required. DSSC’s may be the best fit for all the above. The molecules in DSSC’s can absorb the wavelength of the photon even at a very low intensity. Though these cells have several advantages over the other kind of cells discussed above, these cells are having lot of disadvantages. The major concern for these cells is their thermal resistivity. These cells are not associated with better efficiency compared to others. Hence, these cells are not successfully used on a large scale or for any commercial purpose.

Perovskite Based SCs The crystal structure of the type ABX3 is call perovskite structure (Figure 3) on the name of P.V. Pervoski, Russian mineralogist. The power conversion efficiency of these kinds of metal halides shows an immense development over the last few years. Efficient photon absorption is a result of an appropriate band gap in these cells. Recently, ABX3 type of solar cells has been synthesized with A as an organic monovalent cation like CH3NH3+ and B as Sn (II), or Pb (II) and X as halides. The increased photo current in these cells are ascribed to the presence of mixed cations, which reduce the band gap. The hygroscopic nature and thermal instability are one of the major concerns with the perovskite cells.

a) b) Figure 3. a) Different layers in perovskite cell, b) a typical example of OSCs

Organic SCs Small organic molecules can show photovoltaic effects by absorbing light. These cells are manufactured by fabricating organic molecules through inkjet printing or thermal vapour evaporation techniques which are relatively less expensive processes. As organic solar PV cells avoid the use of toxic

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

83

materials, hence they are quite environmentally friendly. These cells are very thin and transparent; as a result, the cost of preparation is reduced to a significant extent. The fabrication process is also very easy and can be fabricated on flexible material. Though these cells are low-cost, but these cells are lack in efficiency and the efficiency of these cells continuously decreases on exposure to sunlight for a long time.

Multi Junction SCs (Tandem Cells) In multi junction solar cell by using multiple p-n junctions, the range of absorption wavelengths can be increased. Due to stacking (tandem configuration) of the several cells the band gap decreases which is responsible for the increase of cell efficiency. By increasing the number of junctions, the efficiency can be increased up to 86.8% which is much higher than the theoretical efficiency of 31% in single junction cells (Gul et al., 2016). It is noticed that the perovskite cells with tandem arrangement can be 26% more efficient. The use of Ga-As based multi junction solar cells become an environmental issue, as gallium arsenide is toxic and can affect several organs like liver, lungs and can disturb immune system also. Advanced Generation SCsCQDs absorb over a wider range of wavelengths in the visible region. Akin to smaller sized quantum dot (QD) semiconductor particles have distinctly different properties than those of the bulk semi-conductors. The bound structure of the QDs and environmentally friendly nature make them a good alternative for the dye sensitized solar cells (DSSCs) (Margraf et al., 2016). Varying the size of the semi-conductor dots the band gap can be controlled, which affects the efficiency of cell. Due to the presence of defects on the surface of the normal quantum dots, the efficiency of the cell is becoming lowered. To achieve the required photo voltage, CQDs are used as dopant in almost all types of 3rd generation SCs. Recent work by Ping Huang and co-workers observed that the power efficiency of the CdS based solar cells can be increased up to 40% when doped with CQDs. A comparative study of different kinds of SCs in a tabular form is given below (Table 1).

Recent and Advanced Development in SCs It has been well established that SCs showed its astonishing potential in the field of carbon-free green energy. Si-based solar cells proved their presence

84

Biswajit Gayen, Pijush Dutta and Chayan Goswami

with high PCE and these 1st generation SCs are used on a large scale for their excellent efficiency and stability. But these 1st generation solar cells are not affordable for all because these are very costly and cannot work in diffused sunlight and damaged cells containing Si which is harmful in concern of human health. On the other hand, in case of 2nd and 3rd generation, SCs are mostly made of metal and their oxide from rare earth’s elements which make these costly ones and, in some cases, very poisonous metal like Pb, Pd, Hg, Cd and As are used. Hence, we made the solar cell affordable through technological development but failed to make the SCs a completely green energy sector (Wu et al., 2021). Table 1. Comparative studies of very few different types of SCs Type of SCs

Generation

Mono-crystalline Poly-crystalline Amorphous silicon Copper zinc tin sulphide (CZTS) Cadmium telluride based SCs Copper indium gallium Diselenide (CIGS) Dye sensitized (DSSC)

First Second

Maximum PCE (%) 25 19.8 13.8

Advantages

Disadvantages

High efficiency Cost-effective User friendly

Costly and non-stable Poor efficiency Lower efficiency

11.3

Cost-effective

Non-stable

16.5

Cost-effective and good efficiency Commercially successful

Use of toxic metals like Cd

19.9

Third

>33

Perovskite based SCs Multi junction (Tandem cells)

21

Organic solar cells (OSC)

Up to 20%

29.5

Low-cost and absorbs wider range of radiation Higher efficiency Efficiency can be increased by proper combinations Cost-effective

Use of rarely available material like Ga Low life-span

Use of toxic metals like Sn and Pb Expensive

Low lifetime

Use of CQDs in SCs Presently, people are working with a view to making solar cells by minimizing or avoiding any harmful materials. In last few years, a number of reports have

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

85

been published where scientists worked hard to achieve their goal by some fantastic emerging and hybrid technologies. Carbon nano-particle like CQDs, CDs have received much interest to develop green and high-performance PVs as solar-harvesting device. In third generation SCs, TiO2 is used as excellent photo-anode especially in dye sensitized solar cells (DSSCs) with remarkable PCE and costeffectiveness. But it has some serious drawbacks like high band gap energy, low absorption capabilities for visible light and poor hole transfer mechanism. Most of dyes used in DSSCs are toxic in nature, unstable in UV region and very expensive also. In this scenario, CDs play magnificent role which have the ability to absorb the light with broad wavelength range, down conversion of UV light to visible light and most importantly, decreases the bandgap of the active photoanode, TiO2 and help to produce high PCE (Zhou et al., 2018)]. Applications of CDs are not only limited in the field of energy but also explosive detection, chemical sensing, food safety, bio-imaging, drug delivery, photocatalysis etc. They are successful as because of their a) fabricated surfaces b) different functional groups c) tuning of their size d) doping by hetero-atoms e) bio-compatibilities f) easy synthetic methodology g) can be synthesized in large scale from natural resources h) cost-effective i) stable on exploring a broad range of radiation and j) environment-friendly nature. CDs are exhibited varied photo-physical as well as chemical properties by simply varying their structure, size and compositions. CDs are very newcomer one compared to other in the nano-particle family. From the start of the 21st century, nano-particle dominated the research and development field and they are mostly made of metal and metal oxide. Generally, CQDs are very well-known by their identity as carbon dots (CDs) having their size within 0-10 nm. Another member of CQDs is graphene quantum dots (GQDs) which are completely different from CDs. GQDs are made of graphene layer having its size less than 100 nm having less than 10 thick layer graphene. Carbons in GQDs are mainly sp2 hybridized whereas in CDs they are both sp2 and sp3 hybridized. Xu et al., (2004) first reported the CDs from single wall carbon nanotubes (SWCNTs). Two years after an accidental discovery of CDs, Sun et al., in 2006 reported the successful synthesis of stable CQDs. In 2008, Ponomarenko et al., first reported the GQDs followed by an extended work of Xu et al., 2004.

Photophysical Properties of CQDs CQDs bring a revolution in the field of research and development due their outstanding photophysical properties. From the first day of discovery to till

86

Biswajit Gayen, Pijush Dutta and Chayan Goswami

date the magic characters of CDs are in the centre of interest for researchers. CDs obtain from different natural sources by different synthetic methodologies provide ample photo-physical properties. CDs are proved for their outstanding absorption and emission spectral behaviour. CDs exhibit varied types of optical properties like UV-Vis, photoluminescence, phosphorescence and chemiluminescence. Generally, CDs show broad absorption peak in ultra-violet (UV) region (250-350 nm) in addition to that an absorption tailing in the visible region. Due to π−π* electronic transition of C=C bonds present within CDs responsible for peaks at around ~240 nm. Peaks at around ~340 nm are due to n-π* electronic transition of carbonyl moieties). Surface modification, doping of heteroatoms have crucial role for alteration of absorption spectra of CDs as well as their emission spectra. Surface defects present in the CDs are recommended for broad spectral peak. Different functionalities make changes HOMO-LUMO energy levels of CDs, resulting a variation of absorption as well as their emission spectra.

(a)(b) Figure 4. A typical absorption spectra of CD; (a) broad absorption spectrum, (b) absorption spectrum for π−π* and n−π* transitions.

Gayen at al., (2019) reported the varied photo-physical properties and their theoretical explanations in their current review. CDs are very well-known for their fabulous photoluminescent behaviors. In normally, photoluminescent spectra of CDs appear in blue to green region. Similarly, as above, photoluminescent properties of selected CDs can be changed by applying surface engineering on CDs (see Figure 4). These photoluminescent properties have been successfully applied in different application fields like sensing.

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

87

CDs are also popular for their phosphorescence properties. Different CDs anchored with different functionalities on their surface, whereas carbonyl groups present on the surface of CDs generated triplet state which is one of the main criteria to exhibit phosphorescence in CDs. (Figure 5) A number of researchers successfully synthesized phosphorescence active CDs and reported their sensing and imaging applications.

a)

b)

Figure 5. A typical (a) absorption and emission spectra, (b) a photoluminescent active as synthesized CDs.

Another very rarely observed photophysical properties is chemiluminescence. CDs exhibited chemiluminescence properties which are not investigated broadly but it carries wide future prospective. Remarkable progresses have been done in synthetical and application fields but little explanations about photophysical properties have been studied so far on CDs. Above mentioned each photophysical properties can be explained by two main theoretical explanations, a) bandgap transition and b) surface defect model. CDs are considered as quantum dots and quantum confinement effect offers photoluminescent properties of CDs. We know that quantum confinement effect is a size depended property. Hence, tuning of size for CDs varied the spectral behavior. Some theoretical studies showed that emission maxima and corresponding intensities varied with the variation of particle sizes and electron delocalization over the surface of sp2-hybrid cyclic network. Different functionalities and doping are causing reduction of band gaps as a result of quantum confinement effects and offered red-shift photoluminescent.

88

Biswajit Gayen, Pijush Dutta and Chayan Goswami

Surface defect model can also accomplish to explain the photoluminescent properties. Surface defects present within the CDs capture excitations and make ready CDs for emission. Surface defects generate emissive sites within the CDs. A number of articles showed that the higher surface defect leading to generation of higher emissive sites within the CDs. Incorporation of different functionalities, surface oxidation and doping by heteroatom resulting surface defects over the surface of CDs.

Use of CDs as Photoactive Layer in SCs CDs are used as photoactive layer in PVs as they can exhibit outstanding optical phenomenon. Optical properties such as absorption and emission can be regulated by tuning the size of CDs. Furthermore, one can control band gaps by doping technology. Hybrid combination of TiO2 and doped CDs offered outstanding results where frontier resonance electron transfer (FRET) mechanism is followed. It has been observed that doped graphene quantum dots (GCDs) combined with TiO2 absorb much better in longer wavelength as photo anode. TiO2 embedded with N-doped CDs absorbs UV radiation and converts them to visible range by its red-shifting properties and offers stability of the cell. PCE of DSSCs decreases with exposure of UV radiation, DSSCs coupled with N-CDs exhibited 23% decrease of PCE whereas, without N-CDs, PCE decreases 67% after exposing the cell for three weeks. Hence, to avoid toxicity, CD is a better replacement of dyes in Quantum Dots Solar Cells (QDSCs) [Riaz et al., 2019]. Simple CDs or doped CDs are also very successful as counter electrodes in DSSCs. Sulphur and other heteroatoms doped CDs exhibited excellent photo-excitation properties in Pt, RuSe and CoSe counter electrodes. Use of GQDs embedded polypyrrole (PPy) counter electrode provided better PCE compared to without doping GQDs in DSSC (Figure 6). Pt or metal-based counter electrodes made SCs much more expensive, whereas use of PPy/conductive polymers with GQDs makes the SCs more affordable. Rezaei at al. (2019) reported a successful eco-friendly dye-sensitized solar cell by using CDs as photoactive layer in combination with TiO2. Similarly, Perovskite solar cells (PSCs) combined with GQDs, CDs offers better results in terms of efficiency. CDs combined with Au also showed better efficiency and cost effectiveness. Here, CDs have helped to reduce the use of noble metal use as counter electrode and lead to a lower cost of SCs.

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

89

Figure 6. A typical presentation of advanced CDs embedded dye-sensitized SCs model.

Use of CQDs as Hole Transfer Layer in SCs In any kind of PV cell, photo-sensitized materials eject electrons and their separation generates a cell potential by an electron transfer mechanism, whereas hole transfer mechanism also plays an important role in building a better potential of the cell. To improve the efficiency in OSCs, PSCs and other third generation PV cells, an extra hole transfer layer (HTL) is embedded with an active anode. Very commonly used HTLs are poly(3,4ethylenedioxythiphene): poly(styrenesulfonate) (PEDOT:PSS) and when it is anchored with doped GQDs, exhibited excellent performance in SCs (Kim et al., 2022). In spite of lower efficiency, organic solar cells (OSCs) are used due to their low cost and wide variety. In general, polymer having conductive in nature or small organic are used for photo-excitation. Modification of CQDs made HTLs electron rich (n-type) and electron deficient (p-type), which are successfully utilized in OSCs. Barman et al., reported doped CQDs with

90

Biswajit Gayen, Pijush Dutta and Chayan Goswami

Nitrogen and Phosphorus made HTLs n-type whereas Boron made p-type and these are successfully utilized in SCs with better efficiency.

Use of CQDs as Electron Transfer Layer in SCs Electron transfer layers (ETLs) in third generation SCs improve the PCE and the stability same as in HTLs, especially in PSCs and OSCs they showed the outstanding output. ETL works in between active layer and cathode and helps to withdraw the electron from active layer to cathode, resulting in higher cell potential and hence better PCE as ultimate results. Very commonly, materials used as ETLs, such as phenyl-C61-butyric acid methyl ester (PC61BM), TiO2, SnO2, and 2,2′,7,7′-tetrakis (N,N-di-p-methoxyphenyl amine)-9,9′-spiro bifluorene (spiro-OMeTAD) and these all are faced serious drawbacks of interfacial charge recombination and current leakage (Liu et al., 2017). When doped CQDs-ETLs are used, the above said problem is overcome successfully. It is noticed that CQDs embedded ETLs provide improved PCE of nearly about 13-14% in PSCs. Therefore, CDs has brought a revolution in all kinds of SCs by doping in photoactive layer (anode) and electron receiver (cathode). As an overall, CQDs offer better electrode potentials and ultimately cell potential and make PV-SCs, as low cost, better stability and affordable for all nations, especially for economically backward countries. These successes are not only remarkable for science and technological development, but also, they lead us to almost very close to long-days-awaited target to generate green energy having zero carbon footprints. It is our firm believes that in near future CQDs will be explored as a completely green counter electrode with modifications.

Future Prospects In spite of outstanding technological development in different kinds of SCs, further research and development are necessary in some specific issues to achieve better efficiency and low-cost devices. Researchers have to be attended the fabrication of CQDs and other variations SCs to obtain high efficiency, lifespan of SCs, avoid using any trace amount of toxic materials in SCs and low-cost devices. It is needless to mention the future importance of surface modifications of CQDs such as photo-anode, HTLs and ETLs for next

Carbon Nanodots in Photovaltic Cells, a Solar Energy Harvester

91

generation technologies in PV based SCs. Therefore, we can hope the abovementioned development of CQDs will completely replace the expensive, heavy-metal-based QDs and rare-earth metal toward production of low-cost green energy without any scratching in our mother nature.

Conclusion In this short discussion we expressed the most recently developed technologies for existing SCs and new generation SCs toward their affordability, ecofriendliness and cost-effective. Very recent developments in this field have explored the utilization of CDSs and its derivatives as photoactive layer because of profound photophysical characteristics of CQDs. Here, we discussed and summarized the successful use of CQDs, GQDs and others in registered 3rd generation SCs viz. OSCs, PSCs and DSSCs through their extraordinary properties like absorption of light with broad range (visible to UV to near IR), stability of cell from UV light by up conversion, reduce the charge recombination and anchoring as better electron transporter to cathode. CQDs proved that these can serve an alternative to different toxic dyes and metal/metal oxide QDs and make SCs green and less expensive.

Acknowledgment The authors express their thanks to the Greater Kolkata College of Engineering and Management for providing scope of work in their Institution. Authors thank Prof. Avijit Banerji from Calcutta University for their valuable suggestion and constant support.

References Awasthi A, Shukla AK, SR MM, Dondariya C, Shukla KN, Porwal D, Richhariya G. Review on sun tracking technology in solar PV system. Energy Reports (2020) 6: 392– 405. Barman MK, Jana B, Bhattacharyya S, Patra A.. Photophysical properties of doped carbon dots (N, P, and B) and their influence on electron/hole transfer in carbon dots–Nickel (II) phthalocyanine conjugates.The Journal Physical Chemistry C (2014) 118(34): 20034–20041.

92

Biswajit Gayen, Pijush Dutta and Chayan Goswami

Bojec P. Tracking report-September 2022, International Energy Association, https://www.iea.org/. Buitrago E, Novello AM, Meyer T. Third-Generation Solar Cells: Toxicity and Risk of Exposure. Helvetica (2020) 9: 103. Jaganmohan M. World’s largest solar PV power plants worldwide 2021, Statistica 0.5 respectively. Power circuit of a buck-boost converter is shown in Figure 9. During conduction state of the semiconductor switch (S1), the inductor (LO) gets energized from the input d.c. supply (VS). As S1 is turned off, the inductor (LO) transfers its stored energy to the load through the diode (D1). Reversal of load voltage polarity and discontinuous input and output currents are the main drawbacks of buck-boost converter. The voltage gain expression of buck-boost converter is given below. Voltage gain =

VO VS

=

δ

1 −δ

(7)

C′uk Converter A C′uk converter can provide the same voltage gain to that of a buck-boost converter. Power circuit of a C′uk converter is shown in Figure 10. During

conduction state of the semiconductor switch (S1), the inductor (L1) gets energized from the input d.c. supply (VS). At the same time, the link capacitor (C1) transfers its stored energy to the load and energises inductor (L2) through the switch (S1). As S1 is turned off, the stored energy of L1 is transferred to C1 and the load gets energized by the stored energy of L2, both through the diode (D1). Thus, C′uk converter has the advantage of continuous input and output current. But, it also suffers from the problem of reversed load voltage polarity. The voltage gain expression of C′uk converter is given below. Voltage gain =

SEPIC Converter

VO VS

=

δ

1−δ

(8)

The single-ended primary-inductance converter (SEPIC) provides unreversed output voltage polarity with same gain to that of conventional buck-boost converter or C′uk converter. Power circuit of a SEPIC converter is shown in Figure 11. During conduction state of the semiconductor switch (S1), the inductor (L1) gets energized from the input d.c. supply (VS). At the same time, the link capacitor (C1) transfers its stored energy to inductor (L2) through the switch (S1). As S1 is turned off, the inductor (L1) in series with the input d.c.

106

Shib Sankar Saha and Biswamoy Pal

supply (VS) releases its stored energy to the load and C1 through the diode (D1). Simultaneously, the stored energy of L2 is transferred to the load through D1. The input current of SEPIC converter is continuous, but the output current is discontinuous in nature. The voltage gain expression of SEPIC converter is given below. Voltage gain =

Zeta Converter

VO VS

=

δ

1−δ

(9)

The Zeta converter, like SEPIC converter provides unreversed output voltage polarity with same gain to that of conventional buck-boost converter. Zeta converter has discontinuous input current, but the output current is continuous in nature. Power circuit of a SEPIC converter is shown in Figure 12. During conduction state of the semiconductor switch (S1), the inductor (L1) gets energized from the input d.c. supply (VS). At the same time, the link capacitor (C1) in series with the input d.c. supply (VS) releases its stored energy to inductor (L2) and the load through the switch (S1). As S1 is turned off, the stored energy of L1 is transferred to C1 and the stored energy of L2 is transferred to the load, both through the diode (D1). The voltage gain expression of Zeta converter is given below. Voltage gain =

VO VS

Figure 7. Buck converter.

=

δ

1 −δ

(10)

Power Converters for Efficient Control and the Utilization of Solar …

Figure 8. Boost converter.

Figure 9. Buck-boost converter.

Figure 10. C′uk converter.

Figure 11. SEPIC converter.

107

108

Shib Sankar Saha and Biswamoy Pal

Figure 12. ZETA converter.

Isolated DC-DC Converters Isolated dc-dc converters utilize a transformer to provide electrical isolation between the input and the load. These converters do not impose any effect to the input or the load side due to any disturbance or noise created on the other side of the converter. Turns ratio adjustment of the isolation transformer provides another degree of freedom in extending the voltage gain of the converter. However, incorporation of an isolation transformer increases size, weight and cost of the converter. Isolated converters also have other significant issues, like additional power loss in transformers, harmful effects of transformer leakage inductance, core saturation, high voltage spikes across the switches etc.

Flyback Converter Flyback converter is widely used in low power applications due to its simple topological structure, less component count, and reduced control complexity. This converter is derived from the conventional buck-boost converter with the inductor replaced by a coupled inductor (transformer). The other components used in a flyback converter are semiconductor switch (Sm), output diode (DO) and filter capacitor CO as shown in Figure 13. During conduction state of the semiconductor switch (Sm), the supply voltage (VS) is impressed across the primary of the flyback transformer. The secondary induced voltage now reverses the output diode (DO). Thus, the magnetizing inductor gets energized from the input d.c. supply (VS). As Sm is turned off, the polarity of secondary induced voltage gets reversed and hence, the stored magnetizing energy of the flyback transformer is transferred to the load through the output diode (DO). The voltage gain expression of flyback converter is given below.

Power Converters for Efficient Control and the Utilization of Solar …

Voltage gain =

VO VS

=

n.δ

1 −δ

109

(11)

where, n is secondary to primary turns ratio of the transformer.

Forwad Converter

Forward converter is also called as isolated buck converter. Power circuit of the forward converter is shown in Figure 14. During conduction state of the semiconductor switch (Sm), the supply voltage (VS) is impressed across the primary of the transformer and the secondary induced voltage energizes both the load and output inductor (LO) through the diode (DO). As Sm is turned off, the stored energy of LO is transferred to the load through the diode (D1). The voltage gain expression of forward converter is given below. Voltage gain =

VO

Voltage gain =

VO

VS

= n. δ

(12)

= n. δ

(13)

Push-Pull Converter A push pull converter, as shown in Figure 15 uses two active switches in the primary side and four diodes on the secondary side along with an L-C filter. When, S1 is in conduction, supply voltage (VS) is impressed across the upper primary winding and the secondary induced voltage energizesthe the load and the output inductor (LO) through the diodes (D3 and D4). As S1 is turned off, the stored energy of LO is transferred to the load through the rectifier diodes. Similarly, with S2 in conduction, power is transferred to the load and LO through lower half of transformer primary and the rectifier diodes (D1 and D2). As S2 is turned off, the stored energy of LO is transferred to the load as before. The active switches at transformer primary of push-pull converter are operated symmetrically, such that the input line current is drawn in both halves of a switching cycle. Thus, apart from transformer core saturation reduction, this converter enjoys the advantages of steadier input current, less line noise and more efficient operation. The voltage gain expression is as below. VS

Half-Bridge Converter A half bridge converter as shown in Figure 16 uses two capacitors (C1 and C2) to equally divide the supply voltage (VS). The switches (S1 and S2) are driven symmetrically connecting the transformer primary across the capacitors (C1

.

110

Shib Sankar Saha and Biswamoy Pal

and C2) alternately, such that an alternating square wave of peak voltage (VS/2) is impressed across the primary winding. The induced secondary voltage is converted to d.c. output by the centre-tap rectifier. Here also, chance of transformer core saturation is avoided by impressing symmetrical alternating voltage across transformer windings. The voltage gain expression is expressed below. Voltage gain =

VO VS

=

n.δ 2

(14)

Full-Bridge Converter A full bridge converter as shown in Figure 17 is formed by combining two half bridge structures and the transformer primary is connected between the mid points of the two limbs. The diagonal switch combinations (S1-S2 or S3S4) are alternately driven in symmetrical order, such that an alternating square wave of peak voltage (VS) is impressed across the primary winding. The induced secondary voltage is converted to d.c. output by the secondary rectifier. Full-bridge converter is the most widely used isolated dc-dc converter suitable for medium to high power applications. The voltage gain expression of full-bridge converter is expressed below. Voltage gain =

VO VS

= n. δ

Figure 13. Flyback converter.

(15)

Power Converters for Efficient Control and the Utilization of Solar …

Figure 14. Forward converter.

Figure 15. Push-pull converter.

Figure 16. Half-bridge converter.

Figure 17. Full bridge converter.

111

112

Shib Sankar Saha and Biswamoy Pal

Hard-Switching Operation of Converters and Concept of Soft-Switching Power converters are traditionally operated in hard-switched pulse-width modulation (PWM) technique. Hard switching method refers to simple switching of the semiconductor devices in their own ability without any auxiliary circuitry. Typical voltage and current waveforms of hard-switched semiconductor devices are shown in Figure 18. In this type of switching, considerable power loss occurs in the switching devices, due to simultaneous appearance of non-zero voltage and current. Other drawbacks of hard-switching operations are listed below. a) Efficiency degradation due to large power loss during switching transitions. b) High switching current and voltage stress. Large current spike at turnon is experienced due to the discharge of parasitic body capacitance through the device itself and due to diode reverse recovery in some cases. Large voltage stress at turn-off is experienced due to sudden interruption of current flowing through any stray inductances present around the power devices. c) High dv/dt and di/dt creating objectionable interference (EMI & RFI) to communication networks. d) Poor reliability due to large switching stress and switching loss in semiconductor devices. Power converters frequently use dissipative snubbers, comprising of diodes and passive components to shift part of the switching power from the semiconductor device to the dissipative elements of snubber circuits. Thus, stress and power loss of semiconductor devices are greatly reduced improving the reliability, but overall converter efficiency is only partly improved.

Power Converters for Efficient Control and the Utilization of Solar …

113

Figure 18. Hard-switched voltage-current transition of semiconductor device.

Figure 19. Soft-switched voltage-current transition of semiconductor device.

On the other side, modern power converters are demanded to be of compact structure with reduced volume and weight, and simultaneously maintaining high efficiency. Operations at high switching frequency reduces size and weight of magnetic and filter components, thereby improving power density of the converters. But, high frequency operation also leads to large switching loss and consequently efficiency degradation, if operated with conventional hard-switching technique. A breakthrough in power converter development came with the introduction of resonant switching of semiconductor devices. Resonant converters use an L-C resonant tank in series/parallel with the load or other suitable places, which forces the voltage and current waveforms of semiconductor devices to vary sinusoidally. Switching of the devices is done when the device voltage or current passes through zero and thus the switching loss, switching stress and EMI of the converter gets minimized. However, load voltage regulation demands resonant

114

Shib Sankar Saha and Biswamoy Pal

converters to operate at variable switching frequencies, unlike conventional PWM converters, which ultimately increases design complexity of magnetics, capacitor filters and drive circuitry. On the other hand, quasi-resonant or commonly termed soft-switching converters activate the resonant circuits only at switching intervals and normal PWM operation takes place during rest of the switching period. Thus, all the advantages of resonant switching as well as constant frequency PWM operation are enjoyed by soft-switching converters (Mauliket al., 2020; Saha, 2011; Saha, 2021; Pal et al., 2023; Shitoleet al., 2018). Semiconductor devices of soft-switching converters change their states either with zero-voltage across the switches (ZVS) and/or zero-current through the switches (ZCS). In ZVS or ZCS techniques, the switch voltage and/or current is maintained zero at switching transitions. Thus, by avoiding voltage and current overlap at switching transitions, the loss is minimized. Typical voltage and current waveforms of semiconductor devices operated with soft-switching are shown in Figure 19. However, the best switching condition is achieved, if the semiconductor switch is operated under ZV-ZCS condition by simultaneously maintaining zero voltage and zero current at switching transitions and thereby completely eliminating the switching loss.

Improved Performance DC-DC Converters for SPV Power Systems – Case Studies Power converters are primarily used as an interface between SPV sources and load for voltage level matching and extraction of maximum SPV power. Topologies of modern dc-dc converters for any application in SPV power systems are decided on the key factors of voltage gain range, load port isolation, soft-switching transition of semiconductor devices, efficiency, device voltage and current stress etc. Comprehensive analysis of few recently developed improved performance dc-dc converters of both non-isolated and isolated topologies are presented in this section.

Non-Isolated Soft-Switched Buck Converter A non-isolated soft-switched dc-dc buck converter suitable for wide application in SPV power systems has been presented in (Pal and Saha, 2023). The circuit configuration of proposed converter is depicted in Figure 20. The

Power Converters for Efficient Control and the Utilization of Solar …

115

proposed converter, in addition to the primary components of a classical buck converter, uses an auxiliary network of active and passive components. Integral body diodes of semiconductor switches have also been used in the operation of the converter. The main switch of the converter turns on under ZV-ZCS and turn-off under ZVS condition, whereas the auxiliary switch turns on under ZCS and turns off under ZVS condition. The power diodes also recover softly. As the proposed converter enjoys all the advantages of softswitching, it can be operated at high switching frequency reducing the size and weight of the converter.

Non-Isolated Soft-Switched Boost Converter A non-isolated soft-switching dc-dc boost converter has been presented in (Mandal et al., 2020) for extracting maximum power from solar SPV array. The proposed converter as shown in Figure 21 uses an auxiliary network on the basic structure of a classical boost converter. The auxiliary network comprises of a semiconductor switch, a resonant capacitor, a resonant inductor and three fast recovery diodes. The auxiliary network is used to operate the main and the auxiliary switch of the converter under soft-switching conditions. The main switch is turned on under ZV-ZCS condition and it is turned off with ZVS. The auxiliary switch, however, turns on with ZCS and turns off under ZVS condition. Thus, switching stress of the semiconductor devices have been reduced and simultaneously, the efficiency has been improved.

Non-Isolated Buck-Boost Converter with Extended Voltage Gain An extended voltage gain non-isolated buck-boost converter has been proposed in (Sadhukhan et al., 2020) for maximum power extraction of SPV sources. The structure of the proposed dc-dc converter as shown in Figure 22 consists of two sub-circuits. The first sub-circuit is a classical buck-boost converter and the second sub-circuit, placed at the front end of the proposed converter is used to lift the voltage level of the classical buck-boost converter, thereby enhancing the gain of the proposed dc-dc converter as expressed in equation (16). The proposed converter can also work in buck mode with the same voltage conversion ratio of a classical buck-boost converter. The proposed converter can be efficiently used to extract maximum power from

116

Shib Sankar Saha and Biswamoy Pal

SPV sources with wide range of load resistances and under wide variation in solar irradiance and ambient temperature. Voltage gain,

VO δ (2 - δ) = VS (1 - δ) 2

(16)

Figure 20. Non-isolated soft-switched buck converter.

Figure 21. Non-isolated soft-switched boost Converter.

Non-Isolated Soft-Switched High Step-Up Boost Converter Conventional boost converters can’t be practically operated with a duty ratio in excess of 85% in high frequency applications restricting the voltage gain of the converter to 7. Operation with an extremely large duty ratio leads to large conduction loss in power devices, inductors and filter capacitors. Use of voltage extension cell, switched capacitor, switched inductor or coupledinductor is a common practice to achieve high voltage gain in dc-dc boost converters. The non-isolated soft-switched high step-up coupled-inductor boost converter (Pal and Saha, 2018) shown in Figure 23 can be widely used in SPV applications. Voltage gain of the proposed converter, expressed in

Power Converters for Efficient Control and the Utilization of Solar …

117

equation (17), has been enhanced using a coupled-inductor, primary of which is connected in series with the source and the secondary in series with the load. The proposed converter uses an active auxiliary network, in addition to the basic components required by a coupled-inductor boost converter. The main switch of the proposed converter turns on under ZV-ZCS and turns off under ZVS condition. The auxiliary switch turns on under ZCS and turns off nearly under ZVS condition. The boost diode and other auxiliary diodes also recover softly. Voltage gain,

VO 1 δ (17) = + n. VPV 1-δ 1-δ

Where, n is the secondary to primary turns ratio of coupled inductor.

Figure 22. Non-isolated buck-boost converter with extended voltage gain.

Figure 23. Non-isolated soft-switched high step-up boost converter.

118

Shib Sankar Saha and Biswamoy Pal

Isolated Soft-Switched Phase-Shift Full Bridge DC-DC Converter One major problem in SPV power systems is faced due to the formation of large capacitance between the photovoltaic modules and the ground. According to regulatory guidelines, it is mandatory to ground the metal frames of photovoltaic modules for safety reasons. In grid connected SPV power systems with grounded neutral, a closed loop for the common mode leakage current is formed through the interfacing dc-dc converter and the stray capacitance of photovoltaic modules. The best way to prevent the common mode leakage current is to provide galvanic isolation between grounded grid and SPV-panel thereby eliminating the safety hazards. Phase shift full bridge (PSFB) PWM converters with the integration of high frequency transformers can provide galvanic isolation between the SPV source and the load as shown in Figure 24. The leakage inductance of the transformer and parasitic capacitances of the switching devices have been effectively used in this converter to facilitate soft switching operation of the switches. All the semiconductor switches of the PSFB converter turn on and turn off under ZVS condition, thereby enabling high switching frequency operation to increase compactness and simultaneously maintaining high efficiency of the converter. In many SPV systems, the phase-shift of diagonal switch gate pulses has been modulated to extract maximum power from the solar SPV array, through feed-forward control mechanism (Mukherjee et al., 2021;Sadhukhan et al., 2020). The conventional PSFB converter discussed above, experiences unwanted conduction loss in passive mode due to the circulation of transformer primary current through two semiconductor devices. A modified structure, named ZVZCS PSFB converter using a passive clamping circuit in the secondary side as shown in Figure 25 can minimize the passive mode circulating current (Saha et al., 2010). The leading leg switches of this converter turn on and turn off under ZV-ZCS condition, whereas the lagging leg switches operate under ZCS condition. The current stress of the semiconductor switches of this converter is slightly increased due to the presence of the clamping circuit.

Power Converters for Efficient Control and the Utilization of Solar …

119

Figure 24. Isolated soft-switched phase-shift full bridge dc-dc converter.

Figure 25. Isolated ZV-ZCS phase-shift full bridge dc-dc converter.

Isolated Soft-Switched Flyback Converter Flyback converter is a popular choice in low power SPV applications for the advantages of galvanic isolation, simple structure, reduced component count, and simple drive circuitry requirement. However, conventional flyback converters suffer from the drawbacks of excessive switch voltage stress, caused by the leakage inductance of the transformer. The soft-switched flyback converter shown in Figure 26 uses an active energy recovery cell to recover and utilize the trapped leakage energy of the flyback transformer (Saha et al., 2020). The main and the auxiliary switch of the proposed converter turn on under ZCS and turn off under ZVS conditions. The diodes also recover softly. This converter can be efficiently used in MPPT of low power SPV sources.

120

Shib Sankar Saha and Biswamoy Pal

Grid-Tied Battery Integrated SPV Systems Galvanic isolation between SPV arrays and the grid is a mandatory requirement for grid interfaced SPV power systems. Commercial SPV systems typically use two power conversion stages; a non-isolated dc-dc converter followed by a dc-ac inverter for generating a.c. power from SPV generated d.c. power as shown in Figure 27. The dc-dc converter serves two purposes; voltage level matching (step up or step down) depending on SPV voltage and MPPT control of SPV source. In battery integrated SPV systems, the battery bank is connected to the d.c. link through a bidirectional dc-dc converter for charge/discharge control of battery bank. The second stage is an inverter, output of which is synchronized and tied to the utility or micro-grid. The inverter stage is followed by a power frequency transformer, which provides electrical isolation as well as changes a.c. voltage level. The main drawback of conventional SPV system is faced due to the inclusion of power frequency (50 or 60 Hz) transformer, which increases size, weight, cost and power loss of the SPV system. Modern grid-tied SPV systems use a different architecture as shown in Figure 28. Here, the non-isolated converter is replaced by an isolated PSFB converter, which is operated at high switching frequency under soft-switched condition. Thus, a small, light weight and low cost transformer can be used to provide galvanic isolation and simultaneously conduction and switching loss of the SPV system is reduced. However, the battery bank, in most of the SPV systems, is integrated to the d.c. link through a bidirectional dc-dc converter. Recent researches have proposed many topologies of multi-port dc-dc converters for use in battery integrated grid-tied SPV power systems. Amongst them, the three port dc-dc converter suggested in (Sadhukhan et al., 2022) is very attractive for such applications, as it supports the features of MPPT control of solar SPV source, load voltage regulation and charge/discharge control of battery storage. Structure of this converter is shown in Figure 29. It has two non-isolated input ports; one for connecting the solar SPV source and the other for connecting the battery bank. The load port is isolated from the input ports by a high frequency transformer. The converter operates under soft-switched condition for a specified range of input power.

Power Converters for Efficient Control and the Utilization of Solar …

121

Figure 26. Isolated soft-switched flyback converter with active energy recovery cell.

Figure 27. Commercial SPV system with two power conversion stages.

Figure 28. Architecture of modern grid-tied SPV system.

122

Shib Sankar Saha and Biswamoy Pal

Figure 29. Three port dc-dc converter for SPV applications.

Conclusion In the light of maximizing the solar energy utilization, performance of the SPV system needs to be improved with proper selection of the interfacing dc-dc converters. This chapter presents a brief review of the most common nonisolated and isolated dc-dc converter topologies. Recent advances in power electronic studies suggest implementation of high frequency soft-switching techniques in power converters to achieve the unique features of compact structure, large power-to-volume ratio, high energy efficiency, low device stress (voltage and current), high reliability and minimum interference (EMI & RFI) to nearby communication networks. A comprehensive analysis of few recently developed soft-switched dc-dc converters is presented here for use in SPV applications. This review article will therefore be a convenient reference for the researchers and engineers in choosing or developing the most adequate and suitable dc-dc converter topologies for SPV systems.

Disclaimer None.

Power Converters for Efficient Control and the Utilization of Solar …

123

References Araújoa NMFTS, Sousab FJP, Costab FB. Equivalent models for photovoltaic cell - a review. EngenhariaTérmica (Thermal Engineering) (2020) 19(2): 77-98. Freddy TK, Rahim NA, Hew WP, Che HS. Comparison and analysis of single-phase transformerlessgrid-connected PV inverters. IEEE Transactions on Power Electronics (2014) 29(10): 5358-5369. Gubia E, Sanchis P, Ursua A, Lopez J, Marroyo L. Ground currents in single-phase transformerlessphotovoltaicsystems. Progress in Photovoltaics: Research and Applications (2007) 15: 629–650. Libra M, Poulek V,Kouřím P. Temperature changes of I-Vcharacteristics of photovoltaiccells as a consequence of the fermi energylevel shift.Research in Agricultural Engineering(2017) 63(1): 10-15. Mandal DK, Chowdhuri S, Biswas SK, Saha SS. A soft-switching dc-dc boost converter for extracting maximum power from SPV array. IEEE 5th International Conf. on Computing Communication and Automation (ICCCA), India (2020): 363-368. Maulik A, Pal A, Saha SS. A new soft-switched dc-dc buck converter with large step down ratio. IEEE International Conf. on Power Electronics, Drives and Energy Systems (PEDES), India (2020): 1-6. Mondal M, Saha SS. An MPP tracked SPV system integrating single-stage boost inverter and bidirectional battery charger. National Power Electronics Conference (NPEC), India (2021): 1-5. Mukherjee S, Saha SS, Chowdhury S. Design of duty-ratio and phase-shift control circuits for MPPT of SPV source using ZV-ZCS PSFB converters. Devices for Integrated Circuit (DevIC), India (2021): 555-559. Mumtaz F, Yahaya NZ, Meraj ST, Singh B, Kannan R, Ibrahim O.Review on non-isolated dc-dc converters and their control techniques for renewable energy applications. Ain Shams Engineering Journal (2021) 12(4): 3747-3763. Pal A, SahaSS. Novel zero-voltage zero-current transition buck converter with minimal impact of active auxiliary cell on overall dynamics. IEEE Access (2023) 11: 30083023. Pal A, Saha SS. A new soft-switched high Gain Boost Converter. IEEE International Conf. on Power Electronics, Drives and Energy Systems (PEDES), India (2018):1-6. Raghavendra KV, Zeb K, Muthusamy A, Krishna TN, Kumar SV, Kim DH, Kim MS, Cho HG, Kim HJ. A comprehensivereview of dc-dc convertertopologiesand modulation strategieswithrecentadvances in solarphotovoltaic systems. Electronic,(2020)9(1):141. Sadhukhan A, Gayen PK, Saha SS. High performance of dual input H-bridge dc-dc converter in solar application. International Journal of Electronics and Communications (AEÜ), Elsevier (2022): 154. Sadhukhan A, Gayen PK, Saha SS. Maximum power point tracking of SPV array using phase-shifted PWM dc-dc converter. IEEE Calcutta Conference (CALCON), India (2020): 516-520.

124

Shib Sankar Saha and Biswamoy Pal

Sadhukhan A, Saha SS, MajumdarA, Majumdar B.Anextended voltage gain buck-boost converter for maximum photovoltaic power extraction.International Journal of Electronics, Taylor & Francis(2020) 108(8): 1754-1773. Saha SS. Soft-switched high step-up dc-dcboostconverter for distributed generation. International Journal of Power Electronics (2021) 13(1): 112-131. Saha SS. Efficient soft-switched boost converter for fuel cell applications. International Journal of Hydrogen Energy, Elsevier (2011) 36(2): 1710-1719. Saha SS, Chaar EC, Lamont LA. Efficient ZV-ZCS phase shift PWM dc-dc converter interfaced with PV cell for telecommunication applications. IEEE International Conf. EnergyCon, Manama, Bahrin (2010). Saha S, Saha PK, Saha SS. A new soft switchedflybackconverterwith active energy recovery cell. IEEE International Conf. on Power Electronics, Drives and Energy Systems (PEDES), India (2020). Shitole AB, Sathyan S, Suryawanshi HM, Talapur GG, Chaturvedi P. Soft-switched high voltage gain boost-integratedflybackconverterinterfaced single-phase grid-tied inverter for SPV integration. IEEE Transactions on Indusy Applications (2018) 54(1): 482-493. Wuhua L, Xiangning H. Review of non-isolatedhigh-step-updc-dc converters in photovoltaicgrid-connected applications. IEEE Transactions on Industrial Electronics (2011) 58(4): 1239-1250. Yu S, Wang J, Zhang X, Li F. Complete parasitic capacitance model of photovoltaic panel considering the rainwater. Chinese Journal of Electrical Engineering (2017) 3(3): 7784.

Chapter 5

A Particle Swarm Optimization Approach for the Maximum Power Point Tracking of a Grid Connected Shaded Photovoltaic Generator Mouna Ben Smida1, Phd and Anis Sakly2 1Laboratory

of Automatic, Electrical Systems and Environment (LASEE), The National Engineering School of Monastir (ENIM), University of Monastir, Tunisia 2Department of electrical engineering, The National Engineering School of Monastir (ENIM), University of Monastir, Tunisia

Abstract Recently, there has been a great deal of interest in renewable energy sources for electricity generation, particularly for photovoltaic generators (PV). Solar energy is considered as a strongly important energy source. Several studies suppose that more than 45% of the energy in the world will be generated by photovoltaic array. Therefore, it is necessary to concentrate on studies to reduce their application costs and to increase their performance. In order to reach the last aspect, it is important to note that the power generated by a solar photovoltaic panel depends strongly on weather and climate change. Faced with this conflict, it is essential to optimize the performance of PV systems in order to increase their efficiency. Thus, a Maximum Power Point Tracking (MPPT) technique is needed to maximize the produced energy. Besides, a major interest has been given to the study of possible optimal autonomous exploitation of ∗

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

126

Mouna Ben Smida and Anis Sakly the PV source regardless the climatic conditions. In this sense, optimization algorithms are an appropriate tool for solving complex problems in the field of renewable energy systems. However, the developed algorithms are often highly dependent on the precision of the mathematical model relative to the generator and, among other things, climatic conditions of use. Several unconventional approaches to optimization have been developed in literature. These methods, also called “smart,” are based on meta-heuristics. In fact, the use of intelligent techniques is widely used, whether for the modelling, identification or control of systems; thanks to their adaptability to changes in system parameters, and their robustness to perturbations and modelling errors. In this work, the development and analysis of a shaded solar energy system is optimized by an intelligent approach based on particle swarm optimization and compared with conventional methods.

Keywords: photovoltaic generator, shading, MPPT, PSO, grid

Introduction Motivated by the concerns on energy availability and environmental safety, the renewable systems installation has been considerably increased (Lalouni et al., 2009). Their application has been expanded from providing small electronics to large power stations coupled to the grid. Considerable work has been devoted to improve the performance of alternative energy sources. Solar energy is one of these renewable sources. Photovoltaic (PV) energy sources are developed while their applications are increasing, ranging from providing small electronic devices to large power plants connected to medium and lowvoltage grids. However, PV systems face problems improving overall efficiency and maximizing the available power rand. Therefore, in addition to the excellent geographical conditions, it is relevant to have an effective and appropriate Maximum Power Point Tracking (MPPT) algorithm for the photovoltaic system. A field of interest is devoted to the methodological development of these techniques. Several tracking algorithms have been developed in different studies. The Perturb and Observe method (P&O) and the Incremental Conductance method (Inc-Cond), as well as alternatives of those techniques, are widely employed (D’Souza et al., 2010; Sarvi et al., 2022; Zegaoui et al., 2011). The P&O developed in (Singh et al., 2015), is applied in practice owing to its simple implementation. Yet, sudden illumination changes can perturb the system and

A Particle Swarm Optimization Approach …

127

this algorithm can fail to track the maximum power point (MPP). The IncCond technique, proposed in (Mirbagheri et al., 2013), has good accuracy and efficiency. It overcomes some drawbacks of the P&O algorithm. This method performs well, but the response time for finding the MPP is reduced due to the relatively complex computations required by the control algorithm. However, when PV modules are exposed to partial shade conditions (PSC), different problems are encountered (Shaiek et al., 2013; Radwan et al., 2010) In fact, under this condition, the apparition of multiple local peaks modifies the PV characteristics (Faridet al., 2012). In such cases, conventional MPPT algorithms may miss the global maximum and converge to the first extreme encountered that could be a local MPP. In this framework, several researchers have studied the MPP tracking challenge. (Patel and Vivak, 2008; Chtita et al., 2022). Srinivasarao et al., (2021) have introduced an experimental study based on both P&O and Inc Cond algorithms. Despite the importance of the results, they didn’t take into consideration the partial shading conditions (PSC) that would induce the patterns of multiple peaks in the output P–V curve. In this context, numerous scholars have proposed smart MPPT approaches such as fuzzy logic control (FLC), genetic algorithm (GA) (Ben smida and Sakly, 2016), cerebellar model articulation controller (CMAC) (Wu et al., 2010), and ant colony algorithm (ACA) (Krishnan et al., 2020). Although these methods are characterized by their fast convergence and can be used for multi-point tracking, the control processes of fuzzy sets and CMAC are complicated and computation-intensive; thus, they are not easy to implement. The GA and ACA are limited to the maximum power point tracking of photovoltaic module arrays exhibiting single-peak characteristics (Chao et al., 2019). Meteorological data such as solar radiation, ambient temperature and duration of illumination are recognized as reliable variables for renewable energy sources. They play an important role in PV systems. It is therefore necessary to formulate models for forecasting and estimating these data. However, in many cases these data are not available due to the high cost and complexity of the instrumentation required to record them. In (Celik et al., 2013) authors formulated a neural network to predict daily solar radiation, while (Mohandes et al., 2000) used RBF networks to model the monthly mean daily values of global solar radiation on horizontal surfaces and compared the performance of their method with that of a multilayer perceptron (MLP) model and a classical regression model. Several applications have been developed in the literature for prediction, control and simulation of MPPT in PV systems. In fact, in (Ben smida and Sakly, 2016), a model of the photovoltaic system

128

Mouna Ben Smida and Anis Sakly

based on fuzzy logic and its simulation has been developed and the authors in (Liu et al., 2016) have solved the drawbacks of conventional MPPT commands in the case of shaded photovoltaic systems via the use of the genetic algorithm. In (Seyedmahmoudian et al., 2016) authors have developed a stateof-the-art artificial intelligence-based MPPT technique for mitigating partial shading effects on PV systems. (Soufyane et al., 2015) proposed a maximum power point tracking algorithm based on the artificial bee colony method. The main advantages of the proposed intelligent techniques are as follows: • • • •

The stochastic nature of methaheuristic algorithms made a purely system-independent MPPT technique. Using the intelligent technique protect the MPPT, ensuring that it falls into the local maxima during the partial shading conditions. The system is adaptive; as it is initialized by detecting any changes in temperature and irradiance levels. The system is accurate enough to track the global MPP even during acute partial shading conditions.

Among the intelligent techniques, Particle Smart Optimization (PSO) has shown its performance in items of robustness of MPP tracking for various operational conditions, in the case of partial shaded PVG. Within this context and taking into account continuous changes of irradiance and temperature on the PV system under shaded cells, a PSO based MPPT was originally implemented, evaluated and compared under Matlab/Simpower system software. The developed algorithm does not only overcome the common disadvantage of conventional MPPT methods, but also it ensures the robustness of the control. The remaining of this chapter is organized as follows: Section 2 focuses on the characteristics of the photovoltaic cell. Section 3 introduces the model of a photovoltaic generator. The studied system architecture is presented in Section 4. Section 5 explains the application of the proposed intelligent method in the field of solar energy sources and the concept of using PSO method in controlling the MPPT of a shaded photovoltaic generator. The performance of the developed control strategies is discussed in section 6 using simulation results and finally section 7concludes the chapter.

A Particle Swarm Optimization Approach …

129

Photovoltaic Cell Characteristics of the Photovoltaic Cell The equivalent circuit of Figure 1 models the characteristic that translates the behavior of an elementary PV cell. This circuit includes a current source and a diode in parallel. Series Rs and parallel (shunt) Rsh resistors are also added to highlight the dissipative effects at the cell level.

Figure 1. Block diagram of a photovoltaic cell.

The current delivered by the cell is described by the following relationship:

Iph  ID  Ish  Ipv  h.G

(1)

Vpv  R s Ipv Vd  R sh R sh

(2)

Ish 

ID  Is .(exp(

q(Vpv  R s Ipv ) qVd ) 1)  Is .(exp( ) 1) nKT nKT

(3)

Therefore, the electrical equation that models a PV cell is governed by:

Ipv  Iph  Is .(exp(

q(Vpv  R s Ipv ) nKT

) 1) 

(Vpv  R s Ipv ) R sh

(4)

130

Mouna Ben Smida and Anis Sakly

with G: illumination in [Kw/m2]; h: Characteristic constant in [Am2/Kw]; Is: saturation current; q: Charge of the electron (1,6.10-19 C). k: Boltzmann constant (1,38.10-23 J/K). T: Temperature in degrees of Kelvin (°K); n: ideality factor. The characteristics Ipv = f(Vpv) and Ppv = f(Ipv) are represented in the following Figure 2:

Figure 2. Characteristics Power-Voltage (P-V) and Current-Voltage (I-V)of a photovoltaic cell.

In fact, the characteristics of a photovoltaic cell depend on the following parameters: •

A short-circuit current: this is the maximum current delivered by a photovoltaic cell, it is expressed by the following equation:

Icc  Ipv  Is .(exp(

•

nKT

) 1) 

(Vpv  R s Icc ) R sh

(5)

Open circuit voltage: It corresponds to the voltage drop across the terminals of the diode when the current Ip passes through it.

0  Ipv  Is .(exp( •

q(Vpv  R s Icc )

qVco V ) 1)  co nKT R sh

(6)

Maximum power: this is the optimum operating point defined by:

Popt  Vopt .Iopt

(7)

A Particle Swarm Optimization Approach …

•

Form factor: The form factor FF is defined by the ratio between the maximum power supplied by the cell Popt, and the product of the short-circuit current Isc by the open-circuit voltage.

FF  •

131

Popt Icc  Vco

(8)

Conversion efficiency: This is the power conversion efficiency. It is expressed by the ratio between the maximum power and the incident power and given by:

η

Popt Pin

(9)

Influence of Series Resistance Rs The series resistance changes the slope of the Ipv = f(Vpv) characteristic in the part where it behaves as a voltage source. It does not affect the open circuit voltage (Figure 3).

Figure 3. Influence of series resistance on the characteristics of a PV cell.

Effect of Shunt Resistor Rsh The variation of the shunt resistance causes the increase in the slope of the characteristic of the PV cell (Figure 4).

132

Mouna Ben Smida and Anis Sakly

Figure 4. Influence of the shunt resistance on the characteristics of a PV cell.

A Particle Swarm Optimization Approach …

133

Influence of Temperature The increase in temperature influences the value of the optimum power and the open circuit voltage. The characteristics of a PV cell for different temperature values are given by the following Figure 5.

Figure 5. Influence of temperature on the characteristics of a PV cell.

Influence of illumination An increase in the level of sunshine induces a significant rise in the shortcircuit current as shown by the shape of the characteristics of a PV cell (Figure 6).

Figure 6. Influence of sunshine on the characteristics of a PV cell.

134

Mouna Ben Smida and Anis Sakly

Photovoltaic Generator The power supplied by a PV cell is insufficient to supply high-power loads, so it is essential to combine cells in series to increase the output voltage or in parallel to increase the current depending on the targeted applications. Thus for Ns cells in series, and Np branches in parallel, the power supplied by the PV generator is given by:

Ppv  Ns .N p .Ipv .Vpv

(10)

Photovoltaic Generator-Load Connection The connection via a matching stage allows the extraction of the maximum power at all times. This stage guarantees to fix the operating point of the photovoltaic generator independently of that of the load. Various control strategies ensure the extraction of the maximum power available at the terminals of the generator. The techniques generally applied for these control loops are based on the association between the adaptation stage and a command called MPPT (Maximum Power Point Tracking) which performs a permanent search for the PPM.

MPPT Control MPPT control allows GPV operation that guarantees the continuous production of maximum power. Thus, regardless of the variation in weather conditions, the converter control places the system at the maximum operating point. In fact, the study of the MPPT technique is based on the search for the maximum power point (PPM) guaranteeing an adequate adaptation between the generator and the load in order to transfer the maximum power. In the literature, various publications have addressed MPPT controls.

Overall System Architecture The studied system is composed of a 2 kW photovoltaic generator connected to the grid via an electronic power stage. By following the reference voltage

A Particle Swarm Optimization Approach …

135

estimated by the MPPT control device, the DC-DC converter ensures the impedance matching between the generator and the grid. The overall architecture of the system is given by Figure 7.

Figure 7. Block diagram of a partially shaded photovoltaic system.

Modeling of the Photovoltaic Generator The model of a PVG is composed by Np branches in parallel where each branch comprises Ns modules in series is given by:

VpvNsRsIpv Vpv  Ns R s Ipv NsV T 1)  Ipv  I  I (e ph 0 R sh

(11)

136

Mouna Ben Smida and Anis Sakly

10

2 G1=G2=1000W/m ° T1=T2=25 C

8 Ipv (A)

6 4 2 0

50

0

2500

150 200 250 Vpv (V) a) I-V characteristic

100

300

350

G1=G2=1000W/m 2 T =T =25° C 1 2

Ppv (W)

2000 1500 1000 500 0

0

50

100 150 200 250 300 Vpv (V) (b) P-V characteristic

350

Figure 8. The characteristics of the GPV studied under nominal conditions.

The studied photovoltaic system consists of two panels in series, each of which has 5 modules in series and delivers 2000 W under nominal conditions. The P-V and I-V characteristics of the system are given by Figure 8.

A Particle Swarm Optimization Approach …

137

Modeling of the Shaded Photovoltaic Generator

2000

12

1500

9

I(A)

P (W)

Partial shading results when part of the photovoltaic system is exposed to different sunlight. In this case, the electrical behavior of the overall system will depend on the characteristics of each cell and the illumination conditions. In fact, the shaded modules use part of the developed power and behave as a load, which influences the total electricity production and can cause the “hotspot” problem. Therefore, deflection diodes are added to protect the photovoltaic modules against this phenomenon during partial shading states. Partial shading problems act on the operation of the photovoltaic panel and modify its characteristics by the appearance of multiple peaks. The characteristics of the shaded system are given by Figure 9. The illumination received by the second generator influences the global aspect of the system particularly on the second extremum which can be either a Local Maximum (ML) or a Global Maximum (MG) as shown in the figure.

1000

G =700 w/m 2 G =300 w/m 2 1 2

6

3

500

0

G =700 w/m 2 G =1000 w/m 2 1 2

0

50

100

150 200 V(v)

a) I-V characteristic

250

300

350

0

0

50

100

150 200 V(v)

250

300

350

(b) P-V characteristic

Figure 9. The characteristics of partially shaded PVG under different illuminations.

DC–DC Converter The connection of the GPV to the DC bus is guaranteed through a booster chopper controlled by an MPPT control system. This converter corresponds to an adaptation stage between the GPV output voltage and the load [62].

138

Mouna Ben Smida and Anis Sakly

In fact, it is a simple power converter which basically consists of a voltage source, an inductor, a power electronic switch (usually a MOSFET or an IGBT) and a diode. Usually, it also has a filter capacitor to smooth the output. The chopper must bring up and periodically break the connection of the source with the load through the switch to obtain an adjustable continuous output voltage. In order to track the maximum power, the switch must be operated with the corresponding duty cycle via an MPPT control algorithm. The configuration of the studied booster chopper is given by Figure 10.

Figure 10. Configuration of the boost converter.

 is the duty cycle of the converter and defined by the ratio between the conduction time of the transistor and the operating period the following equation:



t T

(12)

In continuous operation, the quantities (Vpvout, Ipvout) are linked to the output quantities (Vpvin, Ipvin) of the GPV by:  Vpvin    Vpvout  1     Ipv (1 ).Ipvin     out

The converter is governed by the following model:

(13)

A Particle Swarm Optimization Approach …

 dVpvout 1   (Ipvin  ILpv )    dt Cpvin     dILpv 1  (Vpvout  (1  )Vpvin )    dt L  pv  

139

(14)

The Maximum power point tracking is provided with an MPPT command.

MPPT Control Based on PSO Algorithm In this work, the MPPT of the studied shaded PV system is based on the PSO technique. In fact, the algorithm tunes the cyclic ratio of the converter by controlling the output voltage of the generator. The cyclic ratio is considered as the particle at each iteration. For the studied algorithm, 4 particles are considered. The initial position of the population is given by the vector below: x = [0.2 0.4 0.6 0.8]×2Voc

(15)

The fitness function to be maximized presents the generated PV power. It is developed as in (16): Fitness = Vpv*I pv

(16)

The flowchart of the proposed PSO algorithm for GMPP tracking is given in Figure 11. Due to the change in operations conditions, the PSO algorithm is modified in order to search the new MPP again by resetting the initial population whenever it detects a variation of solar irradiance, temperature, and load.

140

Mouna Ben Smida and Anis Sakly

Figure 11. Flowchart of GMPP tracking using PSO.

PQ Inverter The inverter interfaces the PV generator with the main utility power system. It behaves as a power controller between the DC-link and the network by assuring the regulation of the flow of active and reactive power injected in the grid. Its configuration is given by Figure.12.

A Particle Swarm Optimization Approach …

141

Figure 12. Configuration of the studied inverter.

The converter’s leg consists of a group of two IGBTs connected with the same phase. The two conditions for the switching variable  k of each leg k are given by: 1,  k   0, 

S1k  1 and S2k  0  S1k  0 and S2k  1 

k  1, 2,3

(17)

As ideal power switches are considered: ∑2𝑗𝑗=1 𝑆𝑆𝑗𝑗𝑗𝑗 = 1 𝑘𝑘 ∈ {1,2,3}

(18)

The inverter’s voltages Via , Vib , Vic are related to the switching states S11 ,

S12 and S13 according to the following equations: 𝑉𝑉𝑉𝑉𝑉𝑉 2 −1 −1 𝑆𝑆11 𝑉𝑉𝑉𝑉𝑉𝑉 �𝑉𝑉𝑉𝑉𝑉𝑉 � = 3 �−1 2 −1� �𝑆𝑆12 � 𝑉𝑉𝑉𝑉𝑉𝑉 −1 −1 2 𝑆𝑆13

(19)

Control of the Line-Side Converter The grid dynamic model is expressed by:  di g   Vgd  Vid  R g i gd  L d  lg i gq   dt    di gq   Vg  Viq  R g i gq  L  lg i gd   dt  q

(20)

142

Mouna Ben Smida and Anis Sakly

where Vid and Viq are the d-q inverter voltage components, Lg and Rg are the grid inductance and resistance, respectively and igd and igq represent the dq grid current components. The active and reactive powers generated by the photovoltaic system are calculated using equations (21) and (22). P

3 Vg i g  Vgq i gq 2 d d



(21)

Q

3 Vg i g  Vgd i gq 2 q d



(22)





The basic structure of PQ inverter controller is shown in Figure 13; two PI controllers are proposed to control the injected power flow. A d-axis PI controller is used to control the active power, and a q-axis PI controller controls the reactive power. The d-axis reference is generally obtained from DC-link voltage controller and the q-axis reference is set to zero to get unity power factor. In fact, in the present study, no reactive power is exchanged, and the total power extracted from the PV generator is injected to the grid.

Figure 13. Block diagram of PQ inverter controller.

A Particle Swarm Optimization Approach …

143

Simulations Results The behavior of the grid connected shaded PVG is illustrated using numerical simulations carried out under Matlab -Simulink.. The studied shaded generator is exposed to a variable illumination illustrated in Figure 14 and Figure 15 shows its generated power. The P-V characteristic of the GPV is illustrated in figure 16. It is clear that the proposed PSO algorithm follows the maximum power point for the different states. Figure 17, Figure 18 and Figure 19 gives thre response of the Dc bus, grid currents and voltages respectively.

Figure 14. Illumination profile.

Figure 15. P-V characteristics for different studied scenarios.

144

Mouna Ben Smida and Anis Sakly

Figure 16. Photovoltaic generated power.

Figure 17. Response of the DC bus voltage.

18. Grid currents response.

Figure

A Particle Swarm Optimization Approach …

145

Figure 19. Grid voltage response.

In this section, the PSO approach described above is applied to control the MPPT of the studied shaded PV generator. It's clear that this method is able to track the GM of the PV characteristic for each illumination variation. For example, in the case of the first interval of time, the overall maximum of the PV characteristic, as indicated in Figure 15, is of the order of 1038W. The response the generated power, given in Figure 16, converges appropriately to this value. The DC bus voltage allure is given in Figure 17. Finally, the currents and the voltages injected into the network, represented respectively, in Figure 18 and Figure 19 are perfectly sinusoidals with a constant frequency equal to 50 Hz which emphasizes the robustness of the grid sides control strategy.

Conclusion Improvements in the functioning of photovoltaic generators are essential in order to make the integration of these renewable energy sources more competitive in the balance sheet of the global energy production systems. In this framework, the main objective of this work focuses on the optimization of the MPPT of a shaded photovoltaic generator by intelligent techniques. Following a bibliographic study on the PV generators technologies, the study has focused on the development of ordering strategies to meet specific requirements during the operation of photovoltaic generator. As metaheuristic methods can have potential when the system is non-linear, a new PSO-based regulation technique is developed in this work whose detailed and

146

Mouna Ben Smida and Anis Sakly

specific knowledge about the system is not required. Subsequently, the phenomenon of partial shading for photovoltaic generators was addressed. This phenomenon is responsible for the appearance of several peaks in the characteristics of the semi-shaded GPV. In order to circumvent the limitations of conventional techniques, an advanced technique based PSO has been proposed and to differentiate between the global maximum and the local maximum. The study stressed the robustness and performance of MPPT commands based on meta-heuristics. The studied PV system was then integrated into a conversion chain coupled to the electrical grid and controlled by power converters. The results obtained allowed us to validate the simulation of the energy behavior of the studied system and to check the performance of the proposed PSO control.

References Ben Smida M, Sakly A. A comparative study of different MPPT methods for gridconnected partially shaded photovoltaic systems. International Journal of Renewable Energy Research (2016) 6(3): 1082-1090. Celik, Ali N., and Tariq Muneer. Neural network based method for conversion of solar radiation data. Energy conversion and management (2013) 67: 117-124. Chao K H, Lai Y J, Chang W C. Development of a stand-alone photovoltaic system considering shaded effect for energy storage and release. Electronics (2019) 8(5): 567. Chtita S, Motahhir S, El Hammoumi A, Chouder A, Benyoucef AS, El Ghzizal A, Derouich A, Abouhawwash M, Askar SS. A novel hybrid GWO–PSO-based maximum power point tracking for photovoltaic systems operating under partial shading conditions. Scientific Reports (2022) 12(1): 1063 D’Souza N S, Lopes LA, Liu X. Comparative study of variable size perturbation and observation maximum power point trackers for PV systems. Electric Power Systems Research (2010) 80(3): 296-305. Farid B, Saida B, Tarek A. Optimisation of a GPV by an Artificial Intelligence Technical. International journal of renewable energy research (2012) 2(4): 730-735. Krishnan G S, Kinattingal S, Simon S P, Nayak PS. MPPT in PV systems using ant colony optimisation with dwindling population. IET Renewable Power Generation (2020) 14(7): 1105-1112. Lalouni S, Rekioua D, Rekioua T, Matagne E. Fuzzy logic control of stand-alone photovoltaic system with battery storage. Journal of power Sources (2009): 899-907. Liu L, Meng X, Liu C. A review of maximum power point tracking methods of PV power system at uniform and partial shading. Renewable and Sustainable Energy Reviews 53 (2016): 1500-1507.

A Particle Swarm Optimization Approach …

147

Mirbagheri, S. Zahra, Saad Mekhilef, and S. Mohsen Mirhassani. MPPT with Inc. Cond method using conventional interleaved boost converter. Energy Procedia (2013) 42: 24-32. Mohandes M K, Balghonaim A, Kassas M, Rehman S, Halawani T O. Use of radial basis functions for estimating monthly mean daily solar radiation. Solar Energy (2000) 68(2): 161-168. Patel H, Agarwal V. Maximum power point tracking scheme for PV systems operating under partially shaded conditions. IEEE transactions on industrial electronics (2008) 55(4): 1689-1698. Radwan H, Abdel-Rahim O, Ahmed M, Orabi M, El-Koussi A A. Two stages maximum power Point tracking algorithm for PV systems operating under partially shaded conditions. Power System Conference, MEPCON (2010). Sarvi M, Azadian A. A comprehensive review and classified comparison of MPPT algorithms in PV systems. Energy Systems (2022) 13(2): 281-320. Seyedmahmoudian M, Horan B, Soon T K, Rahmani R, Oo A M, Mekhilef S, Stojcevski A. State of the art artificial intelligence-based MPPT techniques for mitigating partial shading effects on PV systems–A review. Renewable and Sustainable Energy Reviews (2016) 64: 435-455. Shaiek Y, Smida M B, Sakly A, Mimouni M F. Comparison between conventional methods and GA approach for maximum power point tracking of shaded solar PV generators. Solar energy (2013) 90: 107-122. Singh P, Palwalia D K, Gupta A, Kumar P. Comparison of photovoltaic array maximum power point tracking techniques. Int. Adv. Res. J. Sci. Eng. Technol (2015) 2(1): 401404. Smida M B, Sakly A. Fuzzy pitch angle control for grid connected variable-speed wind turbine system. In 7th IEEE International Renewable Energy Congress (IREC) (2016): 1-6. Soufyane Benyoucef A, Chouder A, Kara K, Silvestre S. Artificial bee colony based algorithm for maximum power point tracking (MPPT) for PV systems operating under partial shaded conditions. Applied Soft Computing (2015) 32: 38-48. Srinivasarao P, Peddakapu K, Mohamed M R, Deepika K K, Sudhakar K. Simulation and experimental design of adaptive-based maximum power point tracking methods for photovoltaic systems. Computers & Electrical Engineering 89 (2021): 106910. Wu M K, Widodo S. Single input cerebellar model articulation controller (CMAC) based maximum power point tracking for photovoltaic system. In International Symposium on Computer, Communication, Control and Automation (3CA) (2010) 5(2): 439-442. Zegaoui A, Aillerie M, Petit P, Sawicki J P, Charles J P, Belarbi A W. Dynamic behaviour of PV generator trackers under irradiation and temperature changes. Solar Energy (2011) 85(11): 2953-2964.

Chapter 6

An FDTD Study on the Broadband Light Absorption Enhancement in Thin Film Solar Cells Using Metal Nanoparticle Arrays Saritha K Nair1,*, PhD and V. K. Shinoj2, PhD 1Department 2Department

of Physics, Mar Athanasius College (Autonomous) Kothamangalam, India of Physics, Union Christian College, Aluva, India

Abstract Research into thin film solar cell design for enhanced light trapping and efficiency is of importance due to the lower cost of these arising from the lower usage of semiconducting material. In this numerical study using the Finite Difference Time Domain method, we are investigating the broadband absorption enhancement due to localized surface plasmon in thin-film solar cells with metal nanoparticle array over the silicon layer. The effect of the material of the nanoparticle, their size and the interparticle spacing on the absorption enhancement is thoroughly analysed. Three different materials were chosen: silver and gold due to their surface plasmon resonances located in the visible range and aluminium due to its abundance. The obtained results were compared for ascertaining the choice of material, size and interparticle spacing for optimal absorption enhancement.

Keywords: thin film solar cell, plasmonic solar cell, metal nanoparticle, surface plasmon, FDTD *

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

150

Saritha K Nair and V. K. Shinoj

Introduction Photovoltaic cells are a source of energy with potential to replace fossil fuels if effective conversion of sunlight to clean electrical power is achieved. Silicon has been the preferred material of choice for solar cells owing to its low cost, abundance on earth, and non-toxicity (Blakers et al., 2013; Li, 2013). First generation solar cells that are currently dominating the solar cell market are typically made from thick crystalline semiconductor wafers which demand about 40 per cent of cost of a solar module (Glunz et al., 2012; Marcus, 2015). In order for large scale implementation, the production cost still needs to be significantly reduced and efficiency substantially increased. Thin film silicon solar cells are a good choice towards reducing cost because of low cost of materials and processing (Sopori, 2003; Shah et al., 2004). In thin film solar cell technology, silicon thin film of thickness in the range 1-2 µm is deposited on cheap substrates such as glass, plastic, or stainless steel. However, these thin film solar cells have ineffective absorbance near bandgap and research aimed at increasing efficiency of these solar cells are still ongoing (Stranks et al., 2015; Yuan et al., 2014; Sarkar et al., 2018). The use of metal nanoparticles could excite localized surface plasmons. The surface plasmon resonance will significantly improve the light absorption, and thereby efficiency of the solar cell. Hence, we aim at the analysis of the absorption enhancement in the absorption layer of the thin film solar cell. Currently, there are various computational tools available to compute the enhanced near fields of noble metal nanoparticles or nanostructures solving Maxwell’s equations such as finite element methods (FEMs), discrete dipole approximation (DDA), T-matrix method, etc. In FEMs, spatial discretization is performed to obtain a numerical solution to the system of differential equations (White, 2000). But the FEM method scales defectively with simulation volume. In DDA, the particle itself is divided into small volumes. Each of the parts is treated as a simple dipole with polarizability depending on the composition of the particle (Draine et al., 1994). The T-matrix method is more suitable for systems with high symmetry (Simspon et al., 2006). The electromagnetic field in objects with complicated configurations can be calculated conveniently by the finite-difference time-domain (FDTD) (Jackson et al., 1999) method employing Maxwell’s equations. Being a direct time and space solution, FDTD offers the user a unique insight into all types of problems in electromagnetics and photonics. In this context, the finitedifference time domain (FDTD) method is implemented here.

An FDTD Study on the Broadband Light Absorption Enhancement …

151

In this perspective, the absorption enhancement of silicon on addition of metal nanoparticles over the silicon layer of a silicon solar cell is systematically investigated numerically via Finite Difference Time Domain method. The enhancement factor with each type of nanoparticles is calculated to study the improvement in silicon absorption of the solar cell on the addition of spherical metal nanoparticles. A comparison is made between solar cells with silver, gold and aluminium nanoparticles over the silicon absorption layer to choose the best among the three for optimal absorption enhancement.

Methodology Schematic of configuration used for the study is shown in Figure 1. The metal nanospheres are deposited periodically on the silicon substrate. Here, d is the diameter of the particle and p is the period or the interparticle spacing. Silver, gold and aluminium are considered in this study due to the localized surface plasmon properties of these metal nanoparticles. The absorbing layer is silicon because of the wide range of application monocrystalline silicon solar cells have. The optical parameters of silicon, silver, gold and aluminium are from Palik et. al. (Palik et al., 1998). The simulations were performed using a commercial FDTD software package, Lumerical FDTD (version 8.22.2072), available from Lumerical Inc. (Lumerical). This photonic simulation software FDTD from Lumerical, uses the Finite-Difference Time-Domain (FDTD) method (Taflove et al., 2005; Gedney et al., 2011; Sullivan, 2013) for solving Maxwell's equations in complex structures. The incident source is a uniform plane wave with a wavelength range 400-1100 nm. The simulation study uses perfectly matched boundary layers as a highly efficient absorbing boundary condition on the upper and bottom boundaries of the computational domain that absorb the reflected and transmitted fields. Two power monitors – one located at the surface of the silicon, another located at the bottom – are used for calculating the power absorbed in the silicon. The quantum efficiency of a solar cell, QE(λ), is defined in terms of amount of power absorbed inside the silicon substrate Pabs(λ) and power incident on it Pin(λ) as follows (Lumerical, 2020): 𝑄𝑄𝑄𝑄(𝜆𝜆) =

𝑃𝑃𝑎𝑎𝑎𝑎𝑎𝑎 (𝜆𝜆) 𝑃𝑃𝑖𝑖𝑖𝑖 (𝜆𝜆)

(1)

152

Saritha K Nair and V. K. Shinoj

Figure 1. Schematic of configuration used for the simulation.

The integrated quantum efficiency, IQE(λ), of the cell is defined as the ratio of number of photons absorbed by the thin film solar cell and the number of photons falling onto it (Lumerical, 2020). 𝐼𝐼𝐼𝐼𝐼𝐼(𝜆𝜆) =

⍁

∫ 𝜆𝜆ℎ𝑐𝑐𝑐𝑐𝑐𝑐(𝜆𝜆)𝐼𝐼𝐴𝐴𝐴𝐴 1.5 (𝜆𝜆)𝑑𝑑𝑑𝑑 ⍁

∫ 𝜆𝜆ℎ𝑐𝑐𝐼𝐼𝐴𝐴𝐴𝐴 1.5 (𝜆𝜆)𝑑𝑑𝑑𝑑

(2)

where IAM 1.5 is the sun spectrum (R.R.D. Center, 2009), h is the Planck constant and c is the speed of light in vacuum. The absorption enhancement spectrum g(λ) and the absorption enhancement factor (G) are defined for comparing the enhancement in light absorption of bare solar cell on the introduction of the metal nanoparticle array. The value of g(λ) can be obtained by taking the ratio of the quantum

An FDTD Study on the Broadband Light Absorption Enhancement …

153

efficiency of nanoparticles arrayed silicon solar cell QEnp(λ) and of bare solar cell QEbare(λ) over the wavelength range considered. Value of G can be attained by taking the ratio of integrated quantum efficiencies of the nanoparticle arrayed silicon solar cell IQEnp and of bare silicon IQEbare. The expressions for g(λ) and G are given as follows (Lumerical, 2020): 𝑔𝑔(𝜆𝜆) =

𝑄𝑄𝐸𝐸𝑛𝑛𝑛𝑛 (𝜆𝜆)

𝑄𝑄𝐸𝐸𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 (𝜆𝜆)

𝐼𝐼𝐼𝐼𝐸𝐸𝑛𝑛𝑛𝑛

𝑎𝑎𝑎𝑎𝑎𝑎 𝐺𝐺 = 𝐼𝐼𝐼𝐼𝐸𝐸

𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏

(3)

Results and Discussions The performance of the nanosphere arrayed plasmonic thin film solar cell was investigated in terms of the enhancement factor and the absorption enhancement spectrum. Figure 2 shows the absorption enhancement factor G for nanoparticle arrayed thin film solar cell as a function of period (inter-particle spacing) for spherical nanoparticles of different sizes for silver. Figure 3 and Figure 4 shows similar curves for aluminium and gold nanoparticles over silicon absorption layer. In all the three figures, the absorption enhancement factor is seen to be dependent on the nanoparticle size, period of nanoparticle array and material of the nanoparticle. We can therefore infer that the nanoparticle size, period of nanoparticle array and material of the nanoparticle influence the absorption of incident light into the solar cell. It can be seen from Figures 2, 3 and 4 that enhanced absorption is obtained with silver nanospheres on the surface of the thin film solar cell compared to aluminium and gold nanospheres. The absorption enhancement with silver nanospheres is clearly evident in Figure 2 where the values are higher for nanoparticle sizes below 200 nm. Furthermore, we can see that the optimal absorption enhancement for silver nanoparticles is obtained for a period of 0.35 µm. Figure 5 shows the absorption enhancement spectrum g(λ) for silver nanoparticle arrayed thin film solar cell for different nanoparticle sizes at a period of 0.35 µm. It is clear that silver nanoparticles provide a broadband absorption enhancement compared with the bare solar cell over a wide wavelength range. The lower value of G for larger nanoparticles can be explained by lower values of g(λ) at lower wavelengths for these nanoparticles which result in lower IQEnp values. Considering Figure 2 and Figure 5, the optimal enhancement in absorption is obtained for a nanoparticle size of 150 nm.

154

Saritha K Nair and V. K. Shinoj

Figure 2. Absorption enhancement factor G for silver nanoparticle arrayed thin film solar cell.

Figure 3. Absorption enhancement factor G for aluminium nanoparticle arrayed thin film solar cell.

An FDTD Study on the Broadband Light Absorption Enhancement …

155

Figure 4. Absorption enhancement factor G for gold nanoparticle arrayed thin film solar cell.

Figure 5. Absorption enhancement spectrum for silver nanoparticle arrayed thin film solar cell.

156

Saritha K Nair and V. K. Shinoj

Figure 6. Visible light absorption profile of (a) solar cell with nanosphere array and (b) bare solar cell.

Figure 6 illustrates the visible light absorption profile at 500 nm. It indicates that the presence of the silver nanoparticle increases the scattering into the silicon layer and enhances the light absorption of the silicon thin film solar cell. This increased scattering into the silicon layer is caused by surface plasmon resonance of the silver nanoparticles. The surface plasmon resonance of the silver nanoparticles will therefore result in significant improvement in efficiency.

Conclusion The absorption enhancements of silicon layer in silicon thin film solar cells with array of nanospheres of different metals (silver, aluminium and gold) over the silicon layer have been simulated using the FDTD method. The results, the absorption enhancement factor and the absorption enhancement spectrum, obtained from the simulation demonstrate that the light absorption is significantly improved because of the localized surface plasmon resonance of the silver nanospheres over the silicon surface for a broad wavelength range compared to the spherical metal nanoparticles of gold and aluminium. The simulation results show that it is important to optimize the period and the nanoparticle size for optimal absorption enhancement. This work may be used as a model for further theoretical study into optimization of performance of thin film solar cells.

An FDTD Study on the Broadband Light Absorption Enhancement …

157

Acknowledgment Reproduced from Saritha K. Nair and V. K. Shinoj, “Enhanced absorption in thin film silicon solar cell using plasmonic nanoparticles: An FDTD study”, AIP Conference Proceedings 2336, 020003 (2021) and Saritha K. Nair and V. K. Shinoj, “Numerical study on the broadband light absorption enhancement in thin-film plasmonic solar cell with silver nanoparticle array”, AIP Conference Proceedings 2263, 040006 (2020), with the permission of AIP Publishing. Saritha K Nair acknowledges the financial assistance by RUSA through minor research project (Order No. 008/2019/MRP/RUSA dated 01/04/2019) and Shinoj V K acknowledges the financial support obtained from the DSTSERB under Early Career Research Scheme (Ref No: ECR/2016/001708).

References Barton J P, Alexander D R, Schaub S A. Theoretical determination of net radiation force and torque for a spherical particle illuminated by a focused laser beam. Journal of Applied Physics, (1989) 66(10): 4594-4602. Blakers A, Zin N, McIntosh K R, Fong K. High efficiency silicon solar cells. Energy Procedia, (2013) 33: 1-10. Center R R D. Reference solar spectral irradiance: ASTM G-173. National Renewable Energy Laboratory (2009). Draine B T, Flatau P J. Discrete-dipole approximation for scattering calculations. Journal of the Optical Society of America A (1994) 11(4):1491–1499. Gedney S D. Introduction to the finite-difference time-domain (FDTD) method for electromagnetics. Synthesis Lectures on Computational Electromagnetics (2011) 6:1250. Glunz S W, Preu R, Biro D. Crystalline silicon solar cells – state-of-the-art and future developments. Comprehensive renewable energy (2012) 1:353-387. Jackson JD, Fox RF. Radiating systems, multipole fields and radiation. Classical Electrodynamics (1999) 3. Li C. Silicon Based Photovoltaic Materials. Eco-and Renewable Energy Materials (2013) 1-23. Lumerical Inc. Lumerical FDTD. https://www.lumerical.com/products/fdtd/. (2018) Lumerical Inc. Solar Cells Methodology. https://apps.lumerical.com/solar_cells_ methodology.html (2018). Marcus A A. Innovations in sustainability. Cambridge University Press; (2015). Palik E D. Handbook of optical constants of solids. Vol 3. Academic Press; (1998).

158

Saritha K Nair and V. K. Shinoj

Sarkar P, Maji B, Manna A, Panda S, Mukhopadhyay A K. Effect of Surface PlasmonBased Improvement in Optical Absorption in Plasmonic Solar Cell. International Journal of Nanoscience (2018) 17:1760028. Shah A V, Schade H, Vanecek M, Meier J, Vallat‐Sauvain E, Wyrsch N, Kroll U, Droz C, Bailat J. Thin film silicon solar cell technology. Progress in Photovoltaics: Research and Applications (2004) 12:113-142. Simpson S H, Hanna S. Numerical calculation of interparticle forces arising inassociation with holographic assembly. Journal of the Optical Society of America A (2006) 23(6):1419–1431. Sopori B. Thin film silicon solar cells. Handbook of Photovoltaic Science and Engineering (2003):307-357. Stranks S D, Nayak P K, Zhang W, Stergiopoulos T, Snaith H J. Formation of thin films of organic-inorganic perovskites for high-efficiency solar cells. AngewandteChemie International Edition (2015) 54:3240-3248. Sullivan D M. Electromagnetic simulation using the FDTD method. John Wiley & Sons; (2013). Taflove A, Hagness S C, Piket-May M. Computational electromagnetics: the finitedifference time-domain method. The Electrical Engineering Handbook (2005) 3: 629670. White D A. Numerical Modeling of Optical Gradient Traps Using the Vector Finite Element Method. Journal of Computational Physics (2000) 159(1):13–37. Yuan Z, Li X, Jing H. Absorption enhancement of thin-film solar cell with rectangular Ag nanoparticles. Journal of Applied Sciences (2014) 14:823-827.

Chapter 7

Photovoltaic Systems and Applications for the Early Diagnosis of Skin Diseases in Humans Using Artificial Intelligence Paryati*

Department Informatics Engineering, UPN “Veteran,” Yogyakarta, Indonesia

Abstract To achieve widespread access to electricity in Indonesia, especially for hospitals, in carrying out its operations in optimizing tools to diagnose patient diseases, especially on Photovoltaic Systems and Applications for Early Diagnosis of Skin Diseases in Humans using Artificial Intelligence. Also, in promoting the use of renewable energy, is a great solution for the benefit of the whole society. Various techniques and methods have been carried out to achieve access to electricity and increase the use of renewable energy in the energy mix in various places and locations for services to the wider community, especially in hospitals to diagnose and detect disease early in patients. Specially, to provide electricity in hospitals in remote and outermost areas in Indonesia, they clearly needed a new approach that proved technically and economically feasible. The conventional approach of placing power plants in small and remote areas is not an option for providing reliable and cost-effective access to electricity. So, to provide access to electricity, a locally available renewable energy source, namely photovoltaic (PV), will provide a promising solution. Not only competitive in terms of technology but also in terms of very affordable costs, when compared to using small diesel, PV-battery Standalone, PV-battery, PV-battery-diesel Hybrid are *

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

160

Paryati common solutions. So that PLN must have the most optimal technology, in distributing electricity to remote area networks. This is related to the assessment of economic technology, so having basic knowledge in understanding the design of various photovoltaic system technologies is very important. To avoid technical problems, this will have a negative impact on photovoltaic system technology. And to understand the most optimal control strategy, when operating PV system technology either standalone or hybrid with other power plants. Human skin consists of various layers, whether it is a protected layer, or a layer that is very susceptible to disease attacks from the outside, such as from the surrounding environment. There are so many discoveries related to skin diseases, where there are many ways bacteria and viruses on the skin, attack humans such as from environmental conditions, climate or temperature even from direct contact with the host or virus parent. Skin diseases suffered by the community today are rapidly spreading, due to lack of information and knowledge about these skin diseases. Therefore, researchers conducted research on the detection and diagnosis of skin diseases in humans using an expert system, with web-based forward and backward chaining methods, accompanied by methods of disease prevention and treatment. This system is made so that patients with skin diseases understand and understand the types of skin diseases they suffer, and this system provides several preventive solutions according to the current level of disease. The system development methodology used is the waterfall methodology or often referred to as a waterfall. The application programs used to create expert system programs in this study are Dreamweaver MX, XAMPP, Adobe Photoshop, MySQL, Opera / Mozilla Firefox and others. The results of this research are in the form of application programs that can help patients and sufferers of skin diseases, to be able to find out the type of skin disease he suffers, as well as get extensive information about the skin disease he suffers, as well as know the treatment techniques and how to prevent it.

Keywords: photovoltaic, diagnose, skin diseases, waterfall, artificial intelligence

Introduction In a Solar Power Plant (PLTS) system, appropriate and mutually sustainable components are needed so that PLTS can operate in hospitals, especially on Photovoltaic Systems and Applications for Early Diagnosis of Skin Diseases in Humans using Artificial Intelligence. As it should be and can produce maximum electrical energy, to optimize hospital operations in diagnosing and

Photovoltaic Systems and Applications for the Early Diagnosis…

161

early detecting a disease in patients, according to the potential of sunlight energy that exists at that location. Human skin consists of an epidermis layer and a dermis layer. The skin functions as a means of excretion due to the presence of sweat glands located in the dermis layer. The epidermis is composed of a horny layer and a Malpighian layer. The corneum is a layer of dead skin, which can peel off and be replaced by new cells. The Malpighian layer consists of a spinous layer and a germinativum layer. The spinosum layer functions to resist friction from the outside. The germ layer contains cells that are actively dividing, replacing the layers of cells in the corneum. At this time many discoveries of skin diseases that attack humans. So with advances in computer technology can help humans in various fields, one of which is an expert system. An expert system is a computer program designed to model problem solving abilities like an expert. With the development of expert systems, expert system applications can be made that can diagnose skin diseases based on the symptoms suffered by patients and how to treat them. This system diagnoses the type of skin disease based on the symptoms experienced. The types of skin diseases found in this system are: Basal Cell Carcinoma, Atopic Dermatitis, Allergic Contact Dermatitis, Berloque Dermatitis, Bateman’s purpura, Acrochordons, Angioma, Seborrheic Keratosis, Tinea pedis, Actinic keratosis, Acanthosis Nigricans. Treatments are carried out based on the identification of skin diseases experienced by users who have been previously researched by skin experts for each skin disease in the form of suggestions, recommendations, and appeals. The methodology used in the development of this expert system is the waterfall method consisting of system engineering, analysis, design, coding, testing and maintenance (Pressman, 2020). Making this system until the testing stage and the program is successfully executed according to its function.

Literature Review Photovoltaic (PV) Systems In a Solar Power Plant (PLTS) system, appropriate and mutually sustainable components are needed so that PLTS can operate in hospitals, especially on Photovoltaic Systems and Applications for Early Diagnosis of Skin Diseases

162

Paryati

in Humans using Artificial Intelligence. As it should be and can produce maximum electrical energy, to optimize hospital operations in diagnosing and early detecting a disease in patients, according to the potential of sunlight energy that exists at that location. There are various kinds of PLTS topologies and configurations, each of which has its own advantages and disadvantages.

PV Components The main components in the PLTS system can be seen in the figure below (Figure 1). The Balance of System consists of a charge controller, Battery Energy Storage System (BESS), PV inverter, mounting system and enclosure box.

Figure 1. PV Components.

Module PV There are two types of Crystalline Silicon (c-Si) solar module technologies that are often used in Indonesia, namely mono-crystalline (mono c-Si) and poly-crystalline (poly c-SI). Monocrystalline modules have the highest efficiency, but their production costs are still higher than poly-crystalline modules.

Expert System Expert systems are computer-based systems that use knowledge, facts and reasoning techniques in solving problems that can usually only be solved by an expert in the field (Kusrini, 2019).

Photovoltaic Systems and Applications for the Early Diagnosis…

163

Expert System Advantages and Disadvantages of Expert System The advantages of expert systems (Kusumadewi, 2018) are: storing the knowledge and expertise of an expert, increasing output and productivity, being able to retrieve and preserving the expertise of experts, being able to access knowledge, being able to work with incomplete and uncertain information, saving time in decision making. Weaknesses of expert systems (Arhami, 2018) are: knowledge cannot always be obtained easily, limited experts and sometimes the approaches possessed by experts are different, difficulties in making high-quality expert systems and require very large costs in development and maintenance, need to be developed, maintained and carefully tested before use.

Figure 2. Expert System Structure.

Expert System Structure The expert system (Figure 2) consists of two main parts, namely: the development environment and the consulting environment (Kusumadewi, 2018).

164

Paryati

Knowledge Base The knowledge base used in this expert system uses Rule-Based Reasoning, namely knowledge is represented by using rules in the form of if-then. Inference Engine There are two approaches to controlling inference in a rulebased expert system (Arhami, 2015), namely: Forward Chaining, reasoning starts from the facts first to test the truth of the hypothesis. Forward tracking looks for facts that match the IF part of the IF-THEN rule. And Backward Chaining, reasoning starts from the hypothesis first and to test the truth of the hypothesis, the facts must be sought. Backtracking looks for facts that match the IF-AND part of the IF-AND-THEN rule. The two inference methods are influenced by three kinds of searches, namely Depthfirst Search, which conducts an in-depth search of the rules from the root node moving downwards to successive deep levels. Breadth-first Search, moving from the root node, the nodes in each level are tested before moving to the next level. Best-first search, works based on a combination of depth-first search and breadth-first search by taking the advantages of both methods. Skin disease Human skin consists of layers of epidermis and dermis. The skin functions as a means of excretion due to the presence of sweat glands located in the dermis layer. The epidermis is composed of a horny layer and a Malpighian layer. The corneum is a layer of dead skin, which can peel off and be replaced by new cells. The Malpighian layer consists of a spinous layer and a germinativum layer. The spinosum layer functions to resist friction from the outside. The germ layer contains cells that are actively dividing, replacing the layers of cells in the corneum. The Malpighian layer contains the pigment melanin, which gives skin its color. This layer contains blood vessels, hair roots, nerve endings, sweat glands, and oil glands. Sweat glands produce sweat. The amount of sweat released can reach 2,000 ml per day, depending on the body’s needs and temperature regulation. Sweat contains water, salt, and urea. Another function as a means of excretion is as an organ receiving stimulation, protection against physical damage, radiation, germs and for regulating body temperature.

Photovoltaic Systems and Applications for the Early Diagnosis…

165

Analysis and Design Performance Analysis of PLTS PV Systems The performance of the Interactive Microgrid PLTS PV system from the PVSyst simulation results will be explained. Alternative software that can be used include SolarPro, PV*Sol Premium, HOMER, and other software. Meteorological data, i.e., irradiance, temperature, rainfall, and wind speed were obtained from Meteonorm 7.3 software with hourly data resolution. The PV system design that has been obtained in the previous sub-chapter along with meteorological data is input to the PVsyst software. The output of the simulation that you want to get is the Performance Ratio (PR) and the annual energy yield.

System Requirements Analysis The subject of this system is to create an expert system to diagnose diseases and how to cure them. This system aims to help a person to know the type of disease suffered and its healing through sulfur baths and traditional ingredients or traditional medicine, as well as information on medicinal ingredients to help the healing process. The recommendation data generated in this system is equipped with the type of disease, symptoms of the disease and the method of healing so that the user can find out the disease suffered and how to treat it. The system will analyze the answers to each question given in order to obtain answers based on the knowledge base contained in this expert system. Before analyzing the answers, the system first gives a number of questions to the user through the interface about the symptoms of the illness. The system will analyze the answers from the user by conducting a tracking process on the knowledge base.

Knowledge Acquisition The knowledge acquisition process is carried out by gathering knowledge about the type of disease accompanied by symptoms, causes and treatment. The knowledge that must be acquired is the symptoms suffered.

166

Paryati

System Planning This expert system application is designed to retrieve and identify overall data regarding the definition of disease, the cause of the disease, treatment, symptoms of the disease. The design stage of this expert system consists of five designs, namely knowledge representation design, inference engine design, data flow diagrams, design database and interface design.

Knowledge Representation Design Representation of knowledge is done to build this application using rulesbased production rules (rule). The rule structure has two parts, namely antecedents and consequents. The conclusion stated in the THEN section is declared true, if the IF section of the system is also true or in accordance with certain rules. The production rules in this system use two tracings, namely Forward Chaining for disease production rules and Backward Chaining for massage therapy production rules. The production rules for implementation are:

Disease Production Rules Diseases discussed in this implementation include Basal Cell Carcinoma, Atopic Dermatitis, Allergic Contact Dermatitis, Berloque Dermatitis, Berloque Dermatitis, Bateman’s purpura, Acrochordons, Angioma, Seborrheic Keratosis, Tinea pedis, Actinic keratosis, Acanthosis Nigricans. The following are the rules of disease production according to the symptoms: 1) The rules of production of Basal Cell Carcinoma disease IF Pain Basal Cell Carcinoma OR Reddened skin OR Itchy skin OR Scaly skin OR Clumping skin THEN Basal Cell Carcinoma Disease.

Photovoltaic Systems and Applications for the Early Diagnosis…

2) The rules of production of atopic dermatitis IF Pain Atopic Dermatitis OR Reddened skin OR Itchy skin OR Scaly skin OR fever OR asthma THEN Disease Atopic Dermatitis. 3) The rules of production of Allergic Contact Dermatitis IF Pain Allergic Contact Dermatitis OR Reddened skin OR Itchy skin OR Blistered skin THEN Allergic Contact Dermatitis. 4) Berloque Dermatitis production rules IF Pain Berloque Dermatitis OR Reddened skin OR Thickened skin OR Skin irritation THEN Berloque Dermatitis Disease. 5) Bateman’s Purpura disease production rules IF Sick of Bateman’s Purpura OR Wrinkled skin OR Spotted skin

167

168

Paryati

OR Scaly skin OR Thin skin THEN Bateman’s Disease Purpura. 6) Acrochordons disease production rules IF Pain Acrochordons OR Reddened skin OR hyperpigmentation OR occurs in skin folds OR warts appear on the skin THEN Disease of Acrochordons. 7) Angioma disease production rules IF Pain Angiomas OR Reddened skin OR Clumping skin THEN Angioma Disease. 8) Seborrheic Keratosis disease production rules IF Pain Seborrheic Keratosis OR Itchy skin OR warts appear on the skin OR smooth surface THEN Seborrheic Keratosis Disease. 9) Tinea Pedis disease production rules IF Pain Tinea Pedis OR Itchy skin

Photovoltaic Systems and Applications for the Early Diagnosis…

169

OR Blistered skin OR Peeling skin OR Chapped skin OR Sore skin OR Swollen skin THEN Disease of Tinea Pedis. 10) Actinic Keratosis disease production rules IF Pain Actinic Kerotosis OR Thickened skin OR Scaly skin OR Spotted skin THEN Actinic Keratosis Disease. 11) Production rules for Acanthosis Nigricans IF Pain Acanthosis Nigricans OR Thickened skin OR Wrinkled skin OR hyperpigmentation OR occurs in skin folds THEN Disease Acanthosis Nigricans.

Therapeutic Production Rules The rule of therapeutic production is used to determine the therapy to be carried out against a disease. If a disease is known, then it can be done with healing therapy.

170

Paryati

Inference Engine Design In this system the inference engine is made by following the rules of forward tracking and backward tracking. The forward tracking process begins by asking questions about the symptoms experienced by the user, and then from the symptoms entered by the user, conclusions will be obtained about the type of disease, the definition of the disease, the cause of the disease and the treatment of the disease. The backward tracking process starts from the discovery of the disease. Based on the disease found, a treatment recommendation is obtained.

Data Flow Chart Level 0 Data Data Flow Chart DAD level 0 represents all system elements with a single process with input and output data indicated by incoming and outgoing arrows respectively. The system built has two external entities, namely admin and user. Admin has the authority to update data, while users can only use this system and are not authorized to update data. Users only enter data into the system, and then the system will provide output to the user. Database Design Database design discusses conceptual data design, data dictionary and data physical design. Data Dictionary To complete the image of the data flow data above as a substitute for attributes, the data dictionary is used as follows: a. Admin (expert): {userID, passID} b. Analysis of Results (analysis_hasil): {id, name, gender, address, occupation, kd_disease, noip, date} c. Symptoms (symptoms): {kd_symptoms, nm_symptoms} d. Disease (disease): {kd_disease, nm_latin, definition, cause, treatment}

Photovoltaic Systems and Applications for the Early Diagnosis…

171

Physical Design The table structure in this system consists of admin table, result analysis table, symptom table and disease table. Interface Design The interface is designed to produce an interactive, attractive interface. This section consists of: header, menu, content and footer. Performance Ratio (PR) and Annual Energy Yield Input : Data meteorology and Design Pv PLTS Hybrida. Outpit : Performance Ratio (PR) and Yearly Energy Year (1st*Year). PR= EY (1 tahun). Kapasitas PV⁄ GlobIncGSTC⁄=2034 MWh/ 2218 kW1730 kWhm21kWm2⁄ =63,19%. CF= EY (1 tahun). Kapasitas PV ×jumlah jam dalam 1 tahun=2034 MWh2218 kW×8760 h=12,71%. The Performance Ratio (PR) is obtained by dividing the quotient between the first year’s energy yield (EY) and the installed PV capacity (PV capacity) in the hospital for diagnosing and detecting early disease in patients with skin diseases. With the quotient between the irradiance on the PV Module in 1 year (GlobInc) and the irradiance on STC conditions (GSTC). Capacity Factor (CF) is obtained by dividing the first year’s energy yield by multiplying the total capacity of the installed PV system. Based on this calculation, the performance of the PLTS PV System from the PVsyst Simulation Results is as follows: Performance Ratio (PR) 63.19% Energy Yield (1st year) 2147 MHz Capacity Factor (CF) 12.71%

Results and Discussion Poly-crystalline modules are also well established in the market share. This is indicated by the fragment majority in the share of PV delivery technology. Seems like shipments of PV modules with poly-crystalline technology account for 54% of total shipments. PV modules worldwide in 2016 and more than

172

Paryati

about 60% of PV module production is a poly-crystalline technology module. This shows that the production capability of poly-crystalline technology manufacturing has been growing rapidly. This technology has been widely implemented in various places around the world. Although poly-crystalline technology has slightly lower efficiencies than mono-crystalline modules, based on price, manufacturing capability, and availability in field, polycrystalline modules are considered a better choice for implementation in Indonesia. PV Technology Delivery (2016) and Production (2000 – 2017) 8 8 Fraunhofer ISE, Photovoltaics Report (Freiburg, 2018). Hint 03 PV Module Prices The price range for PV modules in Indonesia from various vendors is as follows: 1. For TKDN PV Modules is in the range of 0.37–0.43 USD/Wp 2. For Non-TKDN PV Modules it is in the range of 0.23–0.30 USD/W. After carrying out the analysis and design stages, the next stages are: hardware and software implementation at the implementation stage of computer hardware with the following specifications: Intel(R) Atom(TM) CPU N280 @ 1.66GHz, 1 Gb RAM, 160 Gb Harddisk, Intel(R) GMA 950. And the software used is Microsoft Windows Operating System XP Home Edition Version 2002 Service Pack 3, some supporting software are: Apache Web Server 2.2.2, MySQL 5.0.21, PHP 5.1.4, PhpMyAdmin 2.8.1, Opera v.10, Web Editor: Macromedia Dreamweaver 2018, Adobe Photshop CS3, Installing Apache, php MySQL and phpMyAdmin web servers in this implementation using the XAMPP v program package. 1.5.3. Database Implementation The database used is MySQL v. 5.0.21. As for the stage of manufacture using the help of software phpMyAdmin v. 2.8.1 (Figure 3).

Application, Development and Implementation The software used is Macromedia Dreamweaver 2018. The disease list page contains various types of skin diseases. This is a display of the disease list page (Figure 4). The disease symptom page contains the symptoms of a selected type of skin disease. The following is a display of the disease symptoms page (Figure 5). The patient data page contains the patient’s name, gender, address and occupation. The following is a display of the patient data page (Figure 6).

Photovoltaic Systems and Applications for the Early Diagnosis…

173

The consultation page contains a disease diagnosis by asking several questions that must be answered by the user. The following is a display of the consultation page (Figure 7). The results analysis page is in the form of conclusions drawn by the expert system based on the symptoms that have been mentioned by the user. The following is a display of the results analysis page (Figure 8). The login page is a page that only administrators can access for the purposes of updating data, adding data or deleting data.

Figure 3. PHPMyADMIN v interface page. 2.8.1.

Figure 4. Display of disease list page.

Figure 5. Display of disease symptoms page.

Figure 6. Display of patient data page.

Figure 7. Consultation page display.

Figure 8. Display of the results analysis page.

180

Paryati

Conclusion Based on the results of this study, the following conclusions can be drawn: a.

b.

c. d. e. f. g. h.

The design of the Interactive Microgrid PLTS System at the hospital for the purposes of diagnosing and early detection of skin diseases in humans, obtained the results of determining that the main components and system configurations had different capacity values with calculations in the initial design. Then the value or point that should be used for further processing is the value or point after determining the main components and determining the configuration because this capacity value or point has been adjusted to the availability of components on the market. So that these steps and stages can minimize errors in calculating the cost of the next final project. Provide information to the user about the skin disease he suffers (initial diagnosis) based on the symptoms given. Understand and acquire knowledge models for the symptoms of skin diseases. Assist in early identification of skin diseases, through computer processing, so that further treatment of these diseases can be carried out quickly. Provide information on treatments that can be done and how to overcome them. The data contained in the system can be updated or added as needed.

References Arhami M. Basic Concepts of Expert System, Andi Offset, Yogyakarta (2018). Fraunhofer. Photovoltaics Report, Publisher Freiburg ISE (2018). Gumintang M A, Sofyan M F, Sulaeman I. Design and Control of PV Hybrid System in Practice, Deutsche Gesellschaft für InternationaleZusammenarbeit (GIZ) GmbH, GIZ Jakarta (2020). Harahap M. Skin Diseases, Jakarta: Hippokrotes (2019). Ikawati, Z. Pharmacotherapy of Skin System Disease, Science Exchange, Yogyakarta (2021). Kadir A. Basics of Dynamic Web Programming with (JSP) Java Server Pages, Andii Offset, Yogyakarta (2018). Kusrini. Expert System and Application Theory, Andi Offset, Yogyakarta (2019).

Photovoltaic Systems and Applications for Early Diagnosis…

181

Kusumadewi S. Artificial Intelligence (Techniques and Applications), GrahaIlmu, Yogyakarta (2018). Levi D H, Green M A, Hishikawa Y, Dunlop E D, Hohl-Ebinger J, Ho-Baillie A W. ‘Solar Cell Efficiency Tables (Version 51)’, Progress in Photovoltaics: Research and Applications, (2018) 26:3–12. Meadows G, Flint E. Handbook Skin Disease and Owner, Karisma Publishing Group, Batam Center (2016). Nugroho B. Dynamic Web Programming Applications with PHP and MySQL, Gava Media, Yogyakarta (2019). Permana B. Adobe Photoshop 7.0 Practical Guide Series, Elek Media Komputindo, Jakarta (2018). Prasetyo D D. MySQL Server Database Administration, Elex Media Komputindo, Jakarta (2019). Pressman R. Software Engineering Practitioner Approach (Book One, Andi Offset, Yogyakarta (2020). Subronto. Parasitic and Microbial Infectious Diseases in Skin Human. Gadjah Mada University Press, Yogyakarta (2020). Sunyoto, Andi Building the Web with Asynchronous JavaScript & XML Technology, Andi, Yogyakarta (2021). Sutarman. Building Web Applications with PHP and MySQL, GrahaIlmu, Yogyakarta, (2019).

Chapter 8

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels in Dairy Farms with Different Fixed Tilt Angles Antonio José Steidle Neto*, PhD and Daniela de Carvalho Lopes, PhD

Department of Agrarian Sciences, Federal University of São João del-Rei, Sete Lagoas, Minas Gerais, Brazil

Abstract In most agricultural buildings, photovoltaic panels are kept with a fixed orientation and tilt angle throughout the year due to economic reasons. In this study, the performance of rooftop photovoltaic systems in dairy farms was evaluated, considering the electric energy generation potential and different fixed tilt angles. Long-term (> 37 years) meteorological databases and mathematical models were used for predicting daily global radiations on tilted panels in dairy farms nearby municipalities of Minas Gerais State, Brazil. The simulated electrical energy assumed the photovoltaic panels oriented to North (toward the Equator) and fixed tilt angle ranging from 15º to 35º. Greater seasonal differences of electrical energy generated from the photovoltaic systems were verified as latitudes increase, with lowest latitude city showing an average energy generation potential 12% higher in summer when compared with winter, and the highest latitude city presenting a value of 55%. The optimal tilt angle did not vary between the seasons and should be close to the latitude of the dairy farm. The potential for generating electricity is reduced as the difference in tilt from the optimum angle increases. The proposed *

Corresponding Author’s Email: [email protected].

In: Photovoltaic Systems Editors: Sudip Mandal and Pijush Dutta ISBN: 979-8-89113-102-6 © 2023 Nova Science Publishers, Inc.

184

Antonio José Steidle Neto and Daniela de Carvalho Lopes methodology can be adapted to other agricultural buildings, locations, and types of photovoltaic systems.

Keywords: renewable energy, solar radiation, modelling, agricultural buildings

Introduction The world generation of electric power from photovoltaic panels has grown exponentially in recent years. It appears as a clean and sustainable approach for addressing the increasing demand for energy, contributing to reduce the environmental impacts related to climate changes (Domingos and Pereira, 2021). Brazil has high solar incidence availability, leading to variations in electric energy between 1.09 and 2.26 MWh m-2 from 1999 to 2018, equivalent to 1.7 times the average values verified in Germany (Lucena and Holanda, 2022). Despite hydropower is still predominant in Brazil, solar energy increased from 2.5 to 3.8% in the national electric matrix between 2021 and 2022 (MME, 2022). The country has been expanding its contribution to clean energy generation every year. The number of photovoltaic systems installed in residential and agricultural buildings increased considerably, mainly in the Southeast and South regions (Steidle Neto and Lopes, 2021). According to Lucena and Holanda (2022), it is estimated that by 2024 Brazil will have around 887 thousand photovoltaic systems connected to the energy supply network. According to simulations performed by Barbosa et al., (2020), solar power in Brazil tends to reach more than 36% of the entire electrical energy use in 2050. These advances are mainly due to technological evolution, rising capacity of panels, reducing investment costs, energy security policies, and other enterprise cost reductions. Dairy farms present a particular role to play in the progress of the photovoltaic systems. They have potential for producing their own renewable energy, contributing to environmental factors, lowering production costs, and increasing the farm autonomy (Nacer et al., 2016). Brazilian dairy farms are heterogeneously distributed throughout the country, with great diversity in terms of herd sizes and production systems. This leads to different load profiles, which are generally not static over time, even in dairy farms performing similar operations. Specifically, the change in the natural daytime photoperiod during the year causes a variation in electricity consumption for lighting. Also, disinfection and cleaning with water heating present minor

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 185

energy consumption in summer compared with winter due to the air temperatures. The opposite consumption is verified for milk refrigeration and cooling. In addition, the technology of the equipment used in the dairy activities and the features of the photovoltaic components will also affect this variability. Thus, photovoltaic systems may be more suitable for some dairy farms than for others. This behavior was also verified by previous scientific researches (Breen et al., 2020; Teagasc, 2017; Upton et al., 2013). The producers located in Minas Gerais State appear as leading investors in technology for obtaining higher quality milk, especially because they are responsible for most of the Brazilian production for internal consumption and export. Furthermore, they take the advantages of being close to many educational and research institutions focused on dairy and food engineering (Steidle Neto and Lopes, 2021). The electrical energy demanded from dairy farms is greater than from crop production. This occurs due to the technology used, which requires additional machines and equipment for lighting (cow buildings, milk tank room, and milking parlor), milking, milk refrigeration/cooling, and disinfection/cleaning (milk tank room and milking parlor). Additionally, considerable variations in energy consumptions are generally observed among different locations, mainly caused by the regional diversification of farms due to distinct sizes and stocking rates (Bukowski et al., 2021). Thus, the use of photovoltaic systems in dairy farms has been evaluated regarding technical conditions (Steidle Neto and Lopes, 2021; Nacer et al., 2016), economic efficiency (Bukowski et al., 2021; Zhang et al., 2018), and environmental aspects (Bey et al., 2021; Breen et al., 2021). The optimal design of a photovoltaic system is closely associated with the geographic coordinates, tilt angle, orientation, and solar tracking mechanism (Kuo et al., 2018). Grid-connected rooftop photovoltaic systems are most common in residential and agricultural buildings, since they are cost-effective, requiring fewer components and easy installation compared with the standalone and hybrid photovoltaic systems (Steidle Neto and Lopes, 2022; Babatunde et al., 2018). Furthermore, these systems provide on-site solutions, reducing the energy transmission and distribution costs (Duman and Güler, 2020). Le Roux (2016) reported that the majority of these systems are fixed due to their lowest initial investment and maintenance costs. Rooftop photovoltaic panels are often mounted using a tilt angle equal to the roof pitch (not considering the optimal value), which excludes the costs of acquiring and maintaining the support structure of the panels. This affects the overall performance of the photovoltaic system (Mukisa and Zamora, 2022).

186

Antonio José Steidle Neto and Daniela de Carvalho Lopes

Ceramic tiles are the most popular pitched roof covering solution worldwide, due to advantages such as the different architectural styles, high durability, reduced cost, as well as good thermal and acoustic properties (Ramos et al., 2018). Roslan et al., (2016) indicated that ceramic roof tilt angles can vary from 0° to 60°, recommending the 30° optimum value for easy maintenance purposes. The Brazilian Technical Standards Association recommended inclinations between 18° and 22° (ABNT, 1983). Several studies have focused on obtaining the optimal orientation and tilt angle for photovoltaic panels that achieve the highest energy generation (Kuo et al., 2018; Kaddoura et al., 2016; Soulayman and Hammoud, 2016; Bakirci, 2012; Mehleri et al., 2010). Many works have concentrated on different localities, evaluating the specific diurnal, monthly, seasonal, and yearly variations of solar radiation and their effects on the photovoltaic system performances (Hafez et al., 2017). For tropical regions, most studies investigated tilt angles on ground-mounted and flat roofed buildings (Barbón et al., 2022; Teofilo et al., 2021; Khoo et al., 2013), but pitched rooftop photovoltaic systems have been also analyzed (Mukisa and Zamora, 2022). Specifically, in Brazil, Santos and Rüther (2014) evaluated the restrictions and the solar irradiation incidence on different building surfaces in the capital cities, by using different azimuth and tilt angles. Jacobson and Jadhav (2018) estimated the optimal tilt angles for fixed roof mounted photovoltaic panels for Manaus (Amazonas State), Rio de Janeiro (Rio de Janeiro State) and Boa Vista (Roraima State). Cronemberger et al. (2012) studied the solar power generation potential for different surfaces in several Brazilian cities, also applying distinct tilt angles and orientations. However, as can be verified in the results of these scientific researches, there is a considerable variability in the generation of electricity from photovoltaic systems between different locations and tilt angles, mainly justified by the distinct architectural designs of the buildings and because Brazil is a continental country, covering different climates and with latitudes of the cities varying from 5° North to 33° South (Custódio et al., 2022). It is also evident that the results of these studies were influenced by the micrometeorological conditions of each studied site, such as the variation in cloudiness (predicted by sunshine duration), day length, solar radiation, and air temperature throughout the year. Therefore, the applicability of results from one region to another may be questionable and inappropriate. For this, more studies about the effects of tilt angle are demanded, with the purpose of optimizing the best design for the photovoltaic systems. In this study, the performance of rooftop photovoltaic systems in dairy farms located in municipalities of Minas Gerais State (Brazil) was evaluated,

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 187

based on the electric energy generation potential. Daily global radiation on modular panels was estimated, by using different fixed tilt angles and longterm meteorological databases. This work provides an update of the methodology previously described in literature for this purpose, also offering the possibility of optimizing the efficiency of this type of photovoltaic systems, considering the microclimatological and geographical characteristics of specific locations. Additionally, the proposed methodology can be adapted to other agricultural buildings, locations, and types of photovoltaic systems.

Materials and Methods Long-term meteorological databases of sunshine duration were provided by the Brazilian National Institute of Meteorology (INMET) for three municipalities of the Minas Gerais State (Table 1), that have dairy farms nearby. The sunshine duration was quantified by Campbell-Stokes equipment (Negretti and Zambra, London, UK) installed in conventional meteorological stations of each municipality. The World Meteorological Organization (WMO) defines sunshine duration as the length of time during which direct solar radiation is above 120 W m-2 (WMO, 2008). An advantage of sunshine duration measurements is that they are available since the late nineteenth century (Sanchez-Lorenzo et al., 2013), covering a much longer time than global radiation records (Manara et al., 2015; Wild, 2009). The inconsistent data and missing days were identified and excluded from the databases using electronic spreadsheet functions. Also, comparisons with day length data were performed considering the premise that the sunshine duration may have a maximum value equal to the day length of a sunny day. The daily standard deviation and arithmetic average of sunshine durations were obtained for each municipality. Mathematical models were applied for predicting global radiation on tilted panels on a daily scale, adopting the recommendation that photovoltaic panels must be oriented toward true North in the Southern Hemisphere. Additionally, since the most used roof covering in agricultural buildings is ceramic tile and the photovoltaic panels are directly installed over the roofs, the tilt angle simulations ranged from 15º to 35º with progressive increases of 5°. These inclinations in relation to the horizontal plane cover the entire range of variation (18° to 22°) for the most types of ceramic tiles, recommended by the Brazilian Technical Standards Association (ABNT, 1983).

188

Antonio José Steidle Neto and Daniela de Carvalho Lopes

Table 1. Coordinates and long-term databases for the three municipalities where dairy farms are situated Latitude Longitude (°S) (°W) Pedra Azul 16.00 41.28 Sete Lagoas 19.46 44.25 Juiz de Fora 21.76 43.36 * Historical series without inconsistencies and failures. Municipality

Altitude (m) 648.91 732.00 939.96

Database* (years) 37 46 48

Figure 1 shows the flowchart of the variables required to predict the global solar radiation on tilted photovoltaic panels from geographic latitude, sunshine duration, day number of the year, and tilt angle of panels. This flowchart represents the sequence and interaction between the variables associated with the estimates, improving understanding and making the description of the process more visual and intuitive. The details of all equations were described in Steidle Neto and Lopes (2021), Bey et al., (2016), Yadav and Chandel (2013), Iqbal (1983), Collares-Pereira and Rabl (1979), Spencer (1971), Liu and Jordan (1963), and Prescott (1940). As recommended by Allen et al., (1998), it was assumed that on very cloudy days, at least 25% of the global solar radiation intercepted by the top of atmosphere hit the earth’s surface, especially as diffuse radiation. An isotropic model was adopted for predicting global radiation on tilted panels, assuming that the energy of diffuse radiation is uniform over the sky dome (Iqbal, 1983). Also, the ground-reflected radiation on tilted panels was estimated considering an isotropic reflection. In this study, the photovoltaic system was interconnected to the transmission and distribution cable network of the State Company of electrical energy (grid-connected). Thus, the battery storage was not included in the system with the purpose of reducing initial investment and maintenance costs (Ramli et al., 2015). Taking into account this arrangement, the electrical energy generated from the photovoltaic system and not consumed is fed into the power grid of the energy company. The grid-connected photovoltaic systems are less complex, more efficient, and cost-effective, mainly with economic incentives (Lima et al., 2017).

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 189

Figure 1. Variables required for predicting the global solar radiation on tilted panels and the electrical energy produced from the photovoltaic panels.

190

Antonio José Steidle Neto and Daniela de Carvalho Lopes

The grid-connected photovoltaic system is composed of a modular panel set, an inverter DC to AC, and an electricity meter. The polycrystalline type of panel was adopted due to its more advantageous benefit-cost ratio compared to the monocrystalline one. The most of manufacturers affirm that the polycrystalline panels are installed with a 25-year warranty, reaching an efficiency of 90% during the first 10 years and 80% during the subsequent 15 years. The electrical energy produced from the photovoltaic system for each municipality was estimated considering the global solar radiation on tilted panels (Figure 1), and the parameter values presented in Table 2. Table 2. Specific parameters of the grid-connected photovoltaic system Parameter Quantity of photovoltaic panels

Value 100.0

Reference --6 Multinational Modular area of the photovoltaic panel 1.7 m2 companies Average efficiency of the photovoltaic 6 Multinational 15.9%* panel companies Inverter DC to AC efficiency 90.0% Nacer et al., 2015 Efficiency from cable electricity losses 98.0% Freitas, 2008 * Based on 25 different models of polycrystalline panels under standard test conditions (25 °C, 1000 W m-2, and air mass = 1.5).

The number of 100 photovoltaic panels, presented in Table 2, was adopted in this study considering that the total area of 170 m2 is sufficient to generate energy for supplying most of the daily electricity demand of a dairy farm with a herd size of 158 cows (Steidle Neto and Lopes, 2021). The daily potential values of electricity generated from the rooftop photovoltaic system were integrated for each month of the year. Subsequently, the results were analyzed from the variations in the distinct seasons of the year in the Southern Hemisphere: autumn (March 21 - June 21), winter (June 22 September 22), spring (September 23 - December 21), and summer (December 22 - March 20).

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 191

Results Figures 2, Figure 3, and Figure 4 present the variation of the electric energy (minimum and maximum), estimated from 100 photovoltaic panels installed in dairy farms and sunshine durations of the long-term databases, during the seasons of the year for the municipalities of Pedra Azul, Sete Lagoas, and Juiz de Fora, respectively. The values were obtained assuming that the tilt angle ranged from 15º to 35º for rooftop fixed panels. The optimal tilt angles were 15° for the dairy farms situated in Pedra Azul and 20° for those located in Sete Lagoas and Juiz de Fora. The values obtained in this study were the closest to the local geographic latitudes of the municipalities (Table 1) and did not vary between the seasons for the same municipality. On the other hand, the generated electric energy varied depending on the season and geographic latitude.

Figure 2. Minimum and maximum values of the electric energy generation potential of rooftop photovoltaic system during the summer, autumn, winter, and spring for the dairy farms located in Pedra Azul (latitude 16.00°S), Minas Gerais State, Brazil.

When comparing the studied cities, greater seasonal differences of electricity generated from the rooftop photovoltaic system were verified as latitudes increase. This was more evident for summer and winter. The dairy farms situated in Pedra Azul, which present latitude differences of 3.46° and 5.76° from those of Sete Lagoas and Juiz de Fora, showed an average energy generation potential 12% higher in summer when compared with winter. On the other hand, the values for the dairy farms located in Sete Lagoas and Juiz de Fora were 46% and 55%, respectively.

192

Antonio José Steidle Neto and Daniela de Carvalho Lopes

Figure 3. Minimum and maximum values of the electric energy generation potential of rooftop photovoltaic system during the summer, autumn, winter, and spring for the dairy farms located in Sete Lagoas (latitude 19.46°S), Minas Gerais State, Brazil.

Pedra Azul was also the municipality that presented the smallest variations in electricity generations when comparing the optimal and the worst tilt angles, with average energy gains by using the optimal tilt angle of 4.7, 6.5, 12.0, and 12.6% in summer, winter, autumn, and spring, respectively. When observing the results of Sete Lagoas, the seasonal energy gains by varying the tilt angle were 8.9, 11.0, 29.7, and 38.5% for summer, spring, autumn, and winter, respectively. For Juiz de Fora, the results were 15.8, 20.4, 22.8, and 27.4% for summer, spring, autumn, and winter, respectively.

Figure 4. Minimum and maximum values of the electric energy generation potential of rooftop photovoltaic system during the summer, autumn, winter, and spring for the dairy farms located in Juiz de Fora (latitude 21.76°S), Minas Gerais State, Brazil.

Differently from the observed for the seasonal differences in the electric energy generation, best performances were observed for the rooftop

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 193

photovoltaic systems as latitudes decrease. Thus, a larger annual electric energy generation potential was verified for the dairy farms situated in Pedra Azul (48.23 MWh year-1), followed by Sete Lagoas (44.43 MWh year-1), and Juiz de Fora (35.40 MWh year-1). For the dairy farms with the same optimal tilt angle (20°), Sete Lagoas presented an annual electric energy generation 25.5% greater than Juiz de Fora, which can be related with the differences of 0.54° and 1.76° between the optimal tilt angle and the geographic latitude of these farms, respectively. The average consumption of electrical energy was 33.55 MWh year-1, based on estimates of electricity demand for a herd size of approximately 158 cows. The data used for this prediction were obtained directly from dairy farms located in the municipalities evaluated in this study. Thus, all municipalities have the potential to meet the electricity requirement of the farms, also obtaining profits from the State Company by selling energy surplus. Certainly, this will depend on the load profile, the technology of the equipment used in dairy activities, and the technical characteristics of the photovoltaic system. The variations of the daily average global solar radiation on tilted panels for the dairy farms nearby the studied municipalities are presented in Figure 5. The estimates were based on the optimal tilt angles evaluated in this study, and on the long-term databases with 37, 46, and 48 years for Pedra Azul, Sete Lagoas, and Juiz de Fora, respectively (Table 1). Dairy farms located in Pedra Azul resulted in higher daily average global solar radiation values, followed by Sete Lagoas and Juiz de Fora. As expected, the global solar radiation profiles were similar for the three locations, with lower values for the winter months (June - September) and greater values during summer (December March). Regardless of the facility type of the breeding system for milk production (Free-Stall, Compost Barn, or other), it is recommended that photovoltaic panels are installed on the roof of the cow stable and/or milking parlor. It is also important to reinforce that the panels may be oriented to the true North in dairy farms located in the Southern Hemisphere. Alternatively, in the pasturebased dairy systems, the panels can be used with an additional objective of proposing artificial shading of animals (Figure 6), being placed within of the paddock. This will reduce the thermal stress of cows, acting as a clean energy source and reducing CO2 emission, also providing an additional source of income to farmers (Maia et al., 2020). For tropical climate conditions, it is recommended that the artificial shading area be approximately 3 m2 adult animal-1, with height of the bottom edge of the roof plane of 2.5 - 3.0 m. Thus,

194

Antonio José Steidle Neto and Daniela de Carvalho Lopes

for the 170 m2 of photovoltaic panel area considered in this study, the shading area will be sufficient to protect 57 cows.

Figure 5. Daily average variability of the global solar radiation on the tilted panels for the dairy farms located in Pedra Azul, Sete Lagoas, and Juiz de Fora, Minas Gerais State, Brazil.

Figure 6. Optimal tilt angles and orientation of rooftop solar panels in a pasturebased dairy system for the studied Brazilian municipalities.

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 195

Discussion Several studies reported that photovoltaic panels should be tilted to an angle equivalent or close to the local geographic latitude (Mukisa and Zamora, 2022; Babatunde et al., 2018; Kuo et al., 2018; Hafez et al., 2017; Kalogirou, 2014), with the purpose of minimizing the average solar irradiance incidence angle on the solar panel throughout the year. Babatunde et al., (2018) reported that solar irradiance is the major contribution to the performance of photovoltaic systems. In this study the optimal tilt angle did not vary between the seasons for a same farm, probably because Brazil is a low and mid-latitude country, with high availability of solar radiation during all year (Lucena and Holanda, 2022). Nowak (2015) affirmed that for regions relatively close to Equator, such as the studied municipalities, the sun path varies little over the year, reducing seasonal differences that can result in changes on optimal tilt angle value. However, as mentioned by Stanciu and Stanciu (2014), the day lengths are longer in summer, leading to greater incoming solar irradiance and contributing to better performance of the photovoltaic system during this season, which explains the differences observed in the generated electric energy for a same municipality depending on the season and geographic latitude. The more evident differences of electrical energy generation between summer and winter, as well as the variations associated with the geographic latitudes, are in accordance with other authors (Barbón et al., 2022; Hafez et al., 2017). The variations verified between the years of the long-term databases of sunshine duration can also have led to these results. The exploration of spatial and temporal characteristics for sunshine duration can improve the understanding of the solar radiation variability intercepted by the panels and, consequently, of the generation of electricity. When considering the differences of the rooftop photovoltaic system performances due to tilt angle deviations from the optimal value, Barbón et al., (2022) verified expressive energy losses as tilt angle diverges from the optimal value, with greater values associated with higher latitudes, and affecting the electric energy generation. Despite of the differences of annual electric energy generation potential, verified depending on the latitude of the dairy farms, the use of the optimal tilt angle in the three studied municipalities resulted in values that largely meet de demands of the milk producers. Steidle Neto and Lopes (2021) reported that most common milking technologies consume from 0.25 to 0.40 MWh cow−1 year−1 for traditional parlors, but values between 0.18 and 0.91 MWh cow−1

196

Antonio José Steidle Neto and Daniela de Carvalho Lopes

year−1 can be achieved with more sophisticated techniques and equipment. Specifically, for Brazilian dairy farms located in Minas Gerais State with herd sizes varying from 66 to 158 lactating cows, these authors found energy consumptions between 0.30 and 0.35 MWh cow-1 year-1. When evaluating other agricultural consumptions, Bukowski et al., (2021) affirmed that electrical energy demand for the most common households in rural areas is around 3 MWh year-1. When using grid-connected photovoltaic systems, the dairy farms located nearby the studied municipalities should inject into the network of the state energy company more than the electrical energy usually purchased, contributing to the Brazilian agricultural planning for dairy sector development. Additionally, the uses of grid-connected rooftop photovoltaic systems with optimal tilt angles tend to reduce the cost of milk production and the CO2 emissions. The differences in the daily average global solar radiation observed for the three studied locations occurred probably due to their geographic coordinates and the cloudiness conditions obtained from the long-term databases of sunshine duration. The most accentuated variations observed from mid-spring to end of summer (October - March) coincided with the rainy period, when the alternations between days with completely sunny and cloudy periods are more frequent. Based on these results, it can be affirmed that roof pitch should not represent a limitation for the installation of rooftop photovoltaic systems in dairy farms nearby the analyzed cities, as long as it is equivalent or close to the local geographic latitude of the municipality. That is, roof pitches close to the optimal tilt angles can considerably improve the system performance, represented by the total electric energy generated during the year. Additionally, it is important to emphasize that other factors can reduce the photovoltaic energy yield, such as the natural and artificial obstacles that could affect the solar radiation incidence on the panels, the increase in the panel temperature, humidity, and ultra-violet radiation (Yadav and Chandel, 2013; Cronemberger et al., 2012). Higher tilt angles also tend to facilitate that dust deposition on solar panel surfaces falls off within short periods of time (Babatunde et al., 2018). The use of long-term databases ensures more accurate estimates, adequately representing the electrical energy generation from tilted photovoltaic panels for both present and future conditions. Therefore, this contributes to reliable decision making associated with the installation, maintenance, and operation of photovoltaic systems in the dairy farms.

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 197

The methods and mathematical models described in this work can be suitable for residential and agricultural buildings located in other countries. Also, they can be applied for other configurations, as stand-alone and hybrid photovoltaic systems. Future studies should focus on economic analysis of photovoltaic system installation and its impacts for different tilt angles, including energy costs, payback period and attractiveness rate, as well as its cost-benefit ratio. These works can contribute for encouraging the use of photovoltaic systems in dairy farms, but are complex due to all variables required, such as the tax cred-its received and the values paid for each MWh injected or consumed from the external network of the energy company if the system is grid-connected. Other information is the possible differentiated electrical energy fees charged by the power company, the maintenance costs, and the initial investment required by the photovoltaic system.

Conclusion Greater seasonal differences of electrical energy generated from the photovoltaic systems were verified as latitudes increase, with lowest latitude city showing an average energy generation potential 12% higher in summer when compared with winter, and the highest latitude city presenting a value of 55%. On the other hand, best performances were observed for the rooftop photovoltaic systems as latitudes decrease, with larger annual electric energy generation being observed for the lowest latitude city. Corroborating with previous scientific researches, the optimal tilt angle for fixed photovoltaic panels should be close to the latitude value of the dairy farm, not varying between the seasons. The potential for generating electricity is reduced as the difference in tilt from the optimum angle increases. The variability in the electric energy generation potential of the rooftop photovoltaic system will depend on the geographic location in association with micrometeorological conditions (day length, cloudiness, solar radiation, and air temperature). Despite the small differences in latitude (16.00 S to 21.76 S), the geographic coordinates clearly influenced the global radiation intercepted by the tilted panels. The technology and efficiency of the photovoltaic system will also contribute to differences. The methodology described in this study can be suitable for residential and agricultural buildings located in other countries. Also, they can be adapted to other configurations, as stand-alone and hybrid photovoltaic systems.

198

Antonio José Steidle Neto and Daniela de Carvalho Lopes

Acknowledgments The authors would like to thank the National Institute of Meteorology (INMET) of Brazil for providing the long-term weather data. This research received no external funding.

References ABNT. NBR 8039: Project and execution of tiling with French type ceramic tiles. Brazilian Technical Standards Association, Rio de Janeiro, (1983). Allen R G, Pereira L S, Raes D, Smith M. Crop evapotranspiration: guidelines for computing crop water requirements. Irrigation and Drainage Paper, 56. FAO, Rome, 1998. Babatunde A A, Abbasoglu S, Senol M. Analysis of the impact of dust, tilt angle and orientation on performance of PV plants. Renew Sust. Energ. Rev. (2018) 90:10171026.. Bakirci K. General models for optimum tilt angles of solar panels: Turkey case study. Renew Sust. Energ. Rev. (2012) 16:6149-6159. Barbón A, Bayón-Cueli C, Bayón L, Rodríguez-Suanzes C. Analysis of the tilt and azimuth angles of photovoltaic systems in non-ideal positions for urban applications. Appl. Energy (2022) 305:117802. Barbosa J, Dias L P, Simões S G, Seixas J. When is the sun going to shine for the Brazilian energy sector? A story of how modelling affects solar electricity. Renew Energy (2020) 162:1684-1702. Bey M, Hamidat A, Benyoucef B, Nacer T. Viability study of the use of grid connected photovoltaic system in agriculture: case of Algerian dairy farms. Renew Sust. Energ. Rev. (2016)63:333-345. Bey M, Hamidat A, Nacer T. Eco-energetic feasibility study of using grid-connected photovoltaic system in wastewater treatment plant. Energy (2021) 216:119217. Breen M, Murphy M D, Upton J. Assessing the effect of modifying milking routines on dairy farm economic and environmental performance. Agri. Engineering (2021) 3:266–277. Breen M, Upton J, Murphy M D. Photovoltaic systems on dairy farms: financial and renewable multiobjective optimization (FARMOO) analysis. Appl. Energy (2020) 278:115534. Bukowski M, Majewski J, Sobolewska A. Macroeconomic efficiency of photovoltaic energy production in polish farms. Energies (2021) 14(18):5721. Collares-Pereira M, Rabl A. The average distribution of solar radiation - correlations between diffuse and hemispherical and between daily and hourly insolation values. Sol. Energy (1979) 22:155-164.

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 199 Cronemberger J, Caamaño-Martín E, Sánchez S V. Assessing the solar irradiation potential for solar photovoltaic appli-cations in buildings at low latitudes - Making the case for Brazil. Energy Build (2012) 55:264-272. Custódio I, Quevedo T, Melo A P, Rüther R A. Holistic approach for assessing architectural integration quality of solar photovoltaic rooftops and shading devices. Sol. Energy (2022) 237:432-446. Domingos R M A, Pereira F O R. Comparative cost-benefit analysis of the energy efficiency measures and photovoltaic generation in houses of social interest in Brazil. Energy Build (2021) 243:111013. Duman A C, Güler Ö. Economic analysis of grid-connected residential rooftop PV systems in Turkey. Renew Energy (2020) 148:697-711. Freitas S S A. Dimensionamento de sistemas fotovoltaicos. Master DegreeDissertation. Instituto Politécnico de Bragança, Portugal, 2008. Hafez A Z, Soliman A, El-Metwally K A, Ismail L M. Tilt and azimuth angles in solar energy applications - a review. Renew Sust. Energ. Rev. (2017) 77:147-168. Iqbal M. An introduction to solar radiation. 1st ed. Academic Press, Canada, 1983. Jacobson M Z, Jadhav V. World estimates of PV optimal tilt angles and ratios of sunlight incident upon tilted and tracked PV panels relative to horizontal panels. Sol. Energy (2018) 169:55-66. Kaddoura T O, Ramli M A M, Al-Tukuri Y A. On the estimation of the optimum tilt angle of PV panel in Saudi Arabia. Renew Sust. Energ. Rev. (2016) 65:626-634. Kalogirou S A. Solar Energy Engineering: Processes and Systems. 2nd ed. Academic Press, USA, (2014). Khoo Y S, Nobre A, Malhotra R, Yang D, Rüther R, Reindl T, Aberle A G. Optimal orientation and tilt angle for maximizing in-plane solar irradiation for PV applications in Singapore. IEEE J. Photovolt. (2013) 4(2):647-653. Kuo K C, Liao M S, Wang J C, Lee Y C, Huang C K, Chou C Y, Liu C Y, Hsu H H, Chen P H, Jiang J A. Comprehensive assessment of the long-term energy harvest capabilities for PV systems with different tilt angles: case study in Taiwan. Renew Sust. Energ. Rev. (2018) 97:74-89. Le Roux W G. Optimum tilt and azimuth angles for fixed solar collectors in South Africa using measured data. Renew Energy (2016) 96:603-612. Lima L C, Ferreira L A, Lima Moraes F H B. Performance analysis of a grid connected photovoltaic system in northeastern Brazil. Energy Sustain. Dev. (2017) 37:79-85. Liu B Y H, Jordan R C. A rational procedure for predicting the long-term average performance of flat-plate solar-energy collectors. Sol. Energy (1963) 7: 53-74. Lucena J D A Y, Holanda V G B. Solar photovoltaic technology in Brazil. Int. J. Environ. Sustain (2022) 3(1):149-160. Maia A S C, Culhari E A, Fonsêca V F C, Milan H F M, Gebremedhin K G. Photovoltaic panels as shading resources for livestock. J. Clean. Prod. (2020) 258:120551. Manara V, Beltrano M C, Brunetti M, Maugeri M, Sanchez-Lorenzo A, Simolo C, Sorrenti S. Sunshine duration variability and trends in Italy from homogenized instrumental time series (1936-2013). J Geophys Res Atmos (2015) 120:3622–3641.

200

Antonio José Steidle Neto and Daniela de Carvalho Lopes

Mehleri E D, Zervas P I, Sariveis H, Palyvos N C, Markatos N C. Determination of the optimal tilt angle and orientation for solar photovoltaic arrays. Renew Energy (2010) 35:2468-2475. MME. Monthly energy Bulletin - June 2022. Ministry of Mines and Energy: Brazil. Available online: www.gov.br/mme/pt-br/assuntos/secretarias/spe/publicacoes/ boletins-mensais-de-energia (accessed on 11/10/2022). Mukisa N, Zamora R. Optimal tilt angle for solar photovoltaic modules on pitched rooftops: a case of low latitude equatorial region. Sustain. Energy Technol. Assess. (2022) 50:101821. Nacer T, Hamidat A, Nadjemi O, Bey M. Feasibility study of grid connected photovoltaic system in family farms for electricity generation in rural areas. Renew Energy (2016) 96:305-318. Nacer T, Hamidat A, Nadjemi O. Techno-economic impacts analysis of a hybrid grid connected energy system applied for a cattle farm. Energy Procedia (2015) 75:963968. Nowak B J. Testing and validation of an enhancement to photovoltaic electrical model applied to the source code of ESP-r. Master Degree Dissertation. University of Strathclyde, United Kingdom, 2015. Prescott J A. Evaporation from a water surface in relation to solar radiation. Trans R. Soc. S. Aust. (1940) 64:114-118. Ramli M A M, Hiendro A, Sedraoui K, Twaha S. Optimal sizing of grid-connected photovoltaic energy system in Saudi Arabia. Renew Energy (2015) 75:489-495. Ramos R, Silva A, Brito J, Gaspar P L. Methodology for the service life prediction of ceramic claddings in pitched roofs. Constr. Build Mater. (2018) 166:386-399. Roslan Q, Ibrahim S H, Affandi R, Nawi M N M, Baharun A. A literature review on the improvement strategies of passive design for the roofing system of the modern house in a hot and humid climate region. Front. Archit. Res. (2016) 5(1):126-133. Sanchez-Lorenzo A, Calbó J, Wild M, Azorin-Molina C, Sánchez-Romero A. New insights into the history of the Campbell-Stokes sunshine recorder. Weather (2013) 68(12):327-331. Santos I P, Rüther R. Limitations in solar module azimuth and tilt angles in building integrated photovoltaics at low latitude tropical sites in Brazil. Renew Energy (2014) 63:116-124.. Soulayman S, Hammoud M. Optimum tilt angle of solar collectors for building applications in mid-latitude zone. Energy Convers. Manag. (2016) 124:20-28. Spencer J W. Fourier series representation of the position of the Sun. Search (1971) 2:172. Stanciu C, Stanciu D. Optimum tilt angle for flat plate collectors all over the world - a declination dependence formula and comparisons of three solar radiation models. Energy Convers Manag (2014) 81:133-143. Steidle Neto A J, Lopes D C, Nascimento S T. Potential of grid-connected photovoltaic systems in Brazilian dairy farms. Agri. Engineering (2022) 4(1):122-133. Steidle Neto A J, Lopes D C. Technical analysis of photovoltaic energy generation for supplying the electricity demand in Brazilian dairy farms. Environ Dev Sustain 2021 23(2):1355-1370.

A Long-Term Evaluation of Energy Generation from Photovoltaic Panels … 201 Teagasc. Dairy Farm Infrastructure Handbook, 1st ed.; Animal & Grassland Research and Innovation Centre: Fermoy, Ireland, (2017). Teofilo A, Radosevic N, Tao Y, Iringan J, Liu C. Investigating potential rooftop solar energy generated by leased federal airports in Australia: framework and implications. J. Build Eng. (2021) 41:102390. Upton J, Humphreys J, Groot Koerkamp P W G, French P, Dillon P, De Boer I J M. Energy demand on dairy farms in Ireland. J Dairy Sci (2013) 96:6489-6498. Wild M. Global dimming and brightening: a review. J Geophys Res(2009)114:D00D16. WMO. Guide to Meteorological Instruments and Methods of Observation. World Meteorological Organization, Switzerland, (2008). Yadav A K, Chandel S S. Tilt angle optimization to maximize incident solar radiation: a review. Renew Sust Energ Rev (2013) 23:503-513. Zhang C, Campana P E, Yang J, Yu C, Yan J. Economic assessment of photovoltaic water pumping integration with dairy milk production. Energy Convers. Manag. NC (2018) 177:750-764.

About the Editors

Dr. Sudip Mandal received his PhD degree from the Department of Computer Science and Engineering, University of Calcutta, India in 2019. He received his B.Tech. and M.Tech. in Electronics and Communication Engineering from Kalyani Government Engineering College, India in 2009 and 2011, respectively. Currently, he is working as the Assistant Professor of Electronics and Communication Engineering Department, Jalpaiguri Government Engineering College, Jalpaiguri, India. He was the former Head of ECE Department, Global Institute of Management Technology, Krishnagar, India. His current research interests include 5G Antenna, Computational Biology, Tomography, Artificial Intelligence and optimization. He is a member of the IEEE Computational Intelligence Society. He has 50+ publications in different peer-reviewed journals, and in national and international conferences also. Dr. Mandal also published 2 Indian patents so far and authored 2 text Books so far. The author is editorial and review board member of many peer review international journals. He also worked as Editor of Book-Series published by Springer Nature, Willey and Nova Science. Dr. Pijush Dutta received his PhD degree from Mewar University, India in 2022. He received his B.Tech and M.Tech. in 2007 and 2012, respectively. Currently he is working as Assistant professor & Head at the Department of Electronics & Communication Engineering, Greater Kolkata College of Engineering and Management. He also worked in Global Institute of Management & Technology, West Bengal, India. He has more than 11 years of teaching & more than 7 years of research experience. He completed Dr. Dutta's research interests include Sensor & transducer, nonlinear process

204

About the Editors

control system, Mechatronics system & control, optimization algorithm, intelligent system, Internet of Things (IOT), machine learning & Deep learning, etc. He has 14 International & national Patent & has published more than 45 research articles in his credit, published in international & national journal & conference proceedings. He has 6 author book & 7 book chapters under reputed book publishers Elsevier & Springer. Beyond that Dr. Dutta also reviewer of several SCI journals like Soft Computing, Journal of Intelligence System and many more.

Index

A absorption enhancement, 149, 150, 151, 152, 153, 154, 155, 156, 158 agricultural building(s), 183, 184, 185, 187, 197 AI models, 1, 5, 6, 14, 15, 18, 21, 22, 32, 34, 40, 41, 42, 43, 44, 45, 48, 49, 50, 51 artificial intelligence (AI), vii, 1, 2, 5, 6, 7, 14, 15, 16, 18, 21, 22, 23, 32, 34, 40, 41, 42, 43, 44, 45, 47, 48, 49, 50, 51, 52, 56, 57, 58, 60, 128, 146, 147, 159, 160, 162, 181, 203

B broadband absorption, 149, 153 broadband light absorption, vii, 149, 157

C carbon dioxide, 63 carbon dots (CDs), 75, 76, 77, 85, 86, 87, 88, 89, 90, 91, 92, 93 carbon nanodots, vii, 75 cadmium telluride (CdTe), 63, 67, 68, 69, 72, 81, 84, 93 copper indium gallium di-selenide (CIGS), 63, 67, 68, 72, 74, 81, 84 clean energy, 63, 64, 184, 193 climate change, 63, 76, 125, 184 climatic conditions, 126

coal, 63 converter topologies, 96, 104, 122

D dairy farms, 183, 184, 185, 186, 187, 188, 191, 192, 193, 194, 195, 196, 197, 198, 200, 201 dc-dc converters, 95, 96, 97, 102, 104, 108, 114, 120, 122, 123 diagnose, 159, 160, 161, 165 digital transformation, 1, 6, 16, 48, 49, 50 disease prevention, 160 Deep neural learning (DL), 2, 8, 13, 15, 21, 32, 34, 43 doped carbon dots, 75, 91

E electrical energy, vii, 64, 65, 76, 77, 78, 79, 160, 162, 183, 184, 185, 188, 189, 190, 193, 195, 196, 197 electrical power, 63, 102, 150 electricity, vii, 22, 70, 77, 78, 95, 96, 97, 100, 125, 137, 159, 183, 184, 186, 190, 191, 192, 193, 195, 197, 198, 200 energy resources, 63 environment friendly, 76 environmental conditions, 160

206

F finite-difference time domain (FDTD), 149, 150, 151, 156, 157, 158 forecasting, 1, 5, 6, 9, 11, 12, 15, 16, 22, 23, 24, 25, 26, 28, 29, 30, 31, 32, 33, 34, 35, 36, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 127 fossil fuel(s), 63, 76, 77, 150 frontier resonance electron transfer (FRET), 75, 76, 88

G global solar radiation, 53, 55, 127, 188, 189, 190, 193, 194, 196 global warming, 63, 76 graphene quantum dots (GCDs), 77, 85, 88, 92 greenhouse gas emission(s), 63, 96 grid, v, 22, 28, 32, 34, 38, 50, 53, 55, 56, 96, 118, 120, 121, 124, 125, 126, 134, 140, 141, 142, 143, 144, 145, 146, 147, 185, 188, 190, 196, 197, 198, 199, 200

I

Index maximum power point tracking (MPPT), 2, 14, 38, 39, 41, 42, 43, 45, 50, 52, 53, 54, 55, 56, 58, 96, 97, 100, 102, 103, 119, 120, 123, 125, 126, 127, 128, 134, 135, 137, 138, 139, 145, 146, 147 metal nanoparticle, 71, 73, 149, 150, 151, 152, 156 metal nanoparticle array, 149, 152 meteorological databases, 183, 187 ML method, 8, 9, 13, 16, 17, 21, 29, 39, 40, 41 ML model, 7, 8, 13, 16, 17, 21, 28, 29, 32, 33, 37, 38, 40, 41, 48, 49 modelling, 126, 184, 198 multilayer perceptron (MLP), 127

N nano particles, 64 natural gas, 63 neural networks, 2, 8, 15, 28, 52, 55, 57, 58, 60 non-isolated and isolated converters, 96 non-isolated converters, 104 nonlinear dynamics, 1, 5, 6 non-renewable resources, vii

isolated converters, 108

O

L light trapping, 74, 149

oil, 63, 164 optimal tilt angle, 183, 186, 191, 192, 193, 194, 195, 196, 197, 199, 200

M

P

machine learning (ML), vii, 2, 7, 8, 9, 13, 15, 16, 17, 19, 21, 22, 28, 29, 32, 33, 37, 38, 39, 40, 41, 43, 48, 49, 50, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 137, 164, 204 macromolecules, 64 maintenance, 1, 5, 6, 15, 16, 34, 40, 41, 48, 49, 50, 51, 161, 163, 185, 186, 188, 196, 197

particle swarm optimization (PSO), 2, 3, 10, 18, 20, 21, 22, 26, 35, 36, 42, 54, 125, 126, 128, 139, 140, 143, 145, 146 perovskites, 64, 68, 93, 158 photovoltaic (PV) system, vii, 1, 2, 5, 6, 8, 13, 14, 15, 16, 17, 18, 21, 22, 28, 29, 32, 34, 37, 40, 41, 42, 44, 45, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58,

Index 59, 91, 125, 126, 127, 128, 135, 136, 137, 139, 142, 146, 147, 159, 160, 161, 165, 171, 183, 184, 185, 186, 187, 188, 190, 191, 192, 193, 195, 196, 197, 198, 199, 200 photovoltaic array, 52, 53, 54, 55, 57, 58, 125, 147, 200 photovoltaic cell(s), 54, 59, 61, 63, 64, 75, 76, 92, 97, 98, 99, 100, 123, 128, 129, 130, 150 photovoltaic devices, 71, 75, 92 photovoltaic generator, 125, 126, 128, 134, 135, 137, 145 photovoltaic materials, vii, 63, 64, 74, 157 photovoltaic panels, 59, 183, 184, 185, 186, 187, 188, 189, 190, 191, 193, 195, 196, 197, 199 photovoltaic systems, vii, 1, 55, 56, 57, 128, 146, 147, 159, 160, 161, 183, 184, 185, 186, 187, 188, 193, 195, 196, 197, 198, 200 photovoltaic technology, 64, 92, 199 plasmon, 150, 156 plasmonic solar cell, 149, 157, 158 power converters, 95, 96, 97, 102, 112, 113, 114, 122, 146 power plants, 92, 126, 159 PV system(s), vii, 1, 2, 5, 6, 8, 13, 14, 15, 16, 17, 18, 21, 22, 28, 29, 32, 34, 37, 40, 41, 42, 44, 45, 48, 49, 50, 51, 52, 53, 55, 56, 58, 59, 91, 125, 126, 127, 128, 139, 146, 147, 160, 165, 171, 199

R reliability, 1, 6, 21, 34, 41, 96, 112, 122 renewable energy, vii, 55, 56, 57, 58, 59, 76, 78, 96, 123, 125, 127, 145, 146, 147, 157, 159, 184 renewable energy sources, vii, 76, 96, 125, 127, 145 rooftop, 183, 185, 186, 190, 191, 192, 194, 195, 196, 197, 199, 201

207 rooftop photovoltaic systems, 183, 185, 186, 193, 196, 197

S seasonal differences, 183, 191, 192, 195, 197 seasons, 47, 76, 183, 190, 191, 195, 197 semiconductors, 63, 64, 67, 69, 70 shading, 37, 38, 52, 53, 54, 56, 58, 126, 127, 128, 137, 146, 147, 193, 199 silicon, 63, 67, 69, 79, 80, 81, 84, 92, 93, 97, 149, 150, 151, 153, 156, 157, 158, 162 skin disease(s), vii, 159, 160, 161, 164, 171, 172, 180, 181 smart model, 2, 41, 49, 51 smart PV system, 1, 6, 7, 13, 16, 47, 48, 49, 51 soft-switching, 95, 96, 112, 114, 115, 122, 123 solar array, vii, 14, 22, 57 solar cell(s), vii, 18, 19, 54, 56, 60, 61, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 88, 89, 92, 93, 97, 99, 149, 150, 151, 152, 153, 156, 157, 158, 181 solar energy, vii, 5, 8, 14, 47, 48, 49, 50, 51, 52, 53, 56, 57, 59, 63, 64, 65, 71, 73, 74, 75, 76, 77, 78, 92, 93, 95, 96, 122, 125, 126, 128, 147, 184, 199, 201 solar photovoltaic (SPV) source, 95, 96, 97, 100, 101, 102, 103, 114, 115, 118, 119, 120, 123 solar power, 42, 56, 59, 60, 63, 64, 66, 67, 72, 77, 160, 161, 184, 186 solar radiation, 2, 14, 52, 55, 57, 58, 59, 127, 146, 147, 184, 186, 187, 193, 195, 196, 197, 198, 199, 200, 201 solar tracking, vii, 185 sunshine duration, 186, 187, 188, 191, 195, 196, 199

208

Index

surface plasmon, 149, 150, 151, 156, 158 surface plasmon resonances, 149 sustainable green energy, 75

tilt angles, 183, 186, 187, 192, 196, 197, 198, 199, 200

T

value chain, 1, 6, 16, 49

thin film solar cell(s), 63, 67, 68, 74, 93, 149, 150, 152, 153, 154, 155, 156

W

V

waterfall, 160, 161 waterfall methodology, 160