Abstract
Health monitoring and analysis of photovoltaic (PV) systems are critical for optimizing energy efficiency, improving reliability, and extending the operational lifespan of PV power plants. Effective fault detection and monitoring are vital for ensuring the proper functioning and maintenance of these systems. PV power plants operating under fault conditions show significant deviations in current-voltage (I-V) characteristics compared to those under normal conditions. This paper introduces a diagnostic methodology for photovoltaic panels using I-V curves, enhanced by new techniques combining optimization and classification-based artificial intelligence. The research is organized into two key sections. The first section outlines the implementation of a DC/DC buck-boost converter, which is designed to extract and display real-time data from the PV system based on actual (I-V) measurements. The second section focuses on the comprehensive processing of the experimental dataset, where the Harris Hawks Optimization (HHO) algorithm is combined with machine learning methods to identify the most critical features. The HHO algorithm is combined with an advanced machine learning model, XGBoost, to accurately detect faults within the PV system. The proposed HHO-XGBoost algorithm achieves an impressive accuracy of 99.49%, outperforming other classification-based artificial intelligence methods in fault detection. In validation and comparison with previous approaches, the HHO-XGBoost model consistently outperforms established methods such as GADF-ANN, PCA-SVM, PNN, and Fuzzy Logic, achieving an overall accuracy of 98.48%. This outstanding performance confirms the model's effectiveness in accurately diagnosing PV system conditions, further validating its robustness and reliability in fault detection and classification.