Advanced Fault Diagnosis in Photovoltaic Arrays Using a Metropolis-Hastings Convolutional Variational Autoencoder

Abstract

Effective fault diagnosis is critical for maintaining the operational efficiency and reliability of photovoltaic (PV) arrays. While a multitude of artificial intelligence techniques have been applied to detect and diagnose faults in solar panel systems, this research presents a novel fault diagnosis strategy. Our approach leverages a Variational Autoencoder (VAE), further enhanced by Metropolis-Hastings Monte Carlo sampling and a robust residual convolutional neural network architecture. The proposed Metropolis-Hastings Convolutional Variational Autoencoder (MH-CVAE) demonstrates precise identification and categorization of diverse PV faults—including arc faults, MPPT failures, line-to-line, open circuits, degradation, and partial shading—under varied operational influences. This methodology is benchmarked against established machine learning techniques and existing autoencoder models reported in the literature. Comprehensive simulations highlight the MH-CVAE model’s superior performance, attaining a remarkable 99.86% accuracy on simulated test data and surpassing conventional diagnostic approaches. This novel approach significantly expands diagnostic capabilities for PV arrays, thereby offering potential improvements in system dependability and overall operational efficacy. Consequently, the MH-CVAE method shows considerable promise for advancing PV array fault diagnosis, ultimately contributing to the development of more resilient and efficient solar energy infrastructures.