Smart Fault Diagnosis in Grid Connected PV Systems: AI Methods Under Experimental Evaluation

Abstract

This study compares AI-based methods for fault detection and classification in Grid Connected Photovaoltaic Systems (GPVS) using a high-resolution experimental dataset. Seven fault types, categorized by temporal behavior, were injected, and key electrical parameters were recorded. Results show that Decision Tree (DT) achieved the highest accuracy, while Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) offered a better trade-off between accuracy and speed, making them more suitable for real-time applications.