OPTIMIZED MULTITASK MULTI ATTENTION RESIDUAL SHRINKAGE CONVOLUTIONAL NEURAL NETWORK FOR FAULT DETECTION IN SOLAR PHOTOVOLTAIC SYSTEMS

Abstract

The world's population is growing and technological improvements are causing an excess of energy use over supply. It is imperative to move closer to a dependable, affordable, and sustainable renewable energy source in order to meet future energy demands.This article offers an interesting concept for research instructors to comprehendaneffect of faults on solar panels and what research have been completed by various approaches, with a machine learning approach to discover glitches. The investigation also sheds light on how to improve solar panel power efficiency and perform defect correction.In this research work, Optimized Multitask Multi attention Residual Shrinkage Convolutional Neural Networkfor Fault Detection in Solar Photovoltaic Systems (MMRSCNN-FD-SPV)is proposed.Initially, the input data is collected from Fault Detection Dataset in Photovoltaic Farms. Then, the collected data is preprocessed using Generalized Multi-kernel Maximum Correntropy Kalman Filter (GMMCKF) used to clean data. Afterwards, pre-processed data is given for Quantum Conditional Generative Adversarial Network (QCGAN) for effectively detects the various possible burdens that occur in photovoltaicboards.Then, Multitask Multi attention Residual Shrinkage Convolutional Neural Network (MMRSCNN) is used to identifying the fault analysis of photovoltaic panels as Fault-free system (FFS), string fault (SF), string to ground fault (SGF) and string to string fault (SSF). In general, MMRSCNN does not disclose any acceptance of optimization methods for computing the ideal parameters for assuring exact identification of fault diagnosis of photovoltaic panels. Therefore in this work,Chameleon Swarm Optimization Algorithm (CSOA) [26] suggested to optimize weight parameters of MMRSCNN. The suggestedtechnique is implemented and analyzed with help of performance measures such as Exactness, precision, f1-score, sensitivity, specificity, Error rate and Calculation time.Performance of MMRSCNN-FD-SPV approach attains 28.75%, 26.89% and 32.57% higher accuracy; 31.87%, 24.57% and 32.94% specificity and 25.43%, 19.64% and 31.40% higher sensitivity when analyzed through existing techniques like Then performance of MMRSCNN-FD-SPV proposed technique is associated with existing method like an actualestimation on fault discovery in solar panels (EE-FD-SP) [16] and Line-line fault detection and classification for photovoltaic systems using ensemble learning model based on IV characteristics (LLF-CPV-EL) [17],a novel convolutional neural network-based method for fault sorting in photovoltaic arrays  (CNN-FC-PA) respectively