This project develops an intelligent diagnostic system for photovoltaic (PV) panel degradation classification using Electrochemical Impedance Spectroscopy (EIS) data. A Physics-Informed Neural Network (PINN) is designed by integrating physical laws of Constant Phase Element (CPE) behavior directly into the training loss function. Additionally, a custom Penalty-Enhanced Bagging Decision Forest (PEBDF) is proposed with depth and impurity penalties to improve model generalization and interpretability. Both models are trained and evaluated on a synthetic EIS dataset containing four degradation classes: Healthy, Dusty, Microcracked, and Dusty+Microcracked. Feature importance analysis identifies the ten most significant EIS parameters, enabling efficient classification with minimal input. A web application is implemented using the Flask framework, providing user registration, secure login, and an intuitive interface for entering EIS parameters and receiving instant predictions with confidence scores and probability distribution. The system demonstrates high classification accuracy while maintaining physical consistency through the PINN model and strong generalization via the PEBDF ensemble. This work presents a novel hybrid approach combining physics-guided deep learning and constrained ensemble methods for accurate PV health monitoring.

Keywords: Photovoltaic degradation, Electrochemical Impedance Spectroscopy, Physics-Informed Neural Network, PINN, PEBDF, Flask, EIS classification, PV health monitoring, Constant Phase Element, Machine learning