A predictive machine learning approach for early fault identification in solar photovoltaic systems

The reliability of photovoltaic (PV) panels plays a pivotal role in optimizing energy output, particularly as solar energy becomes an indispensable pillar in global initiatives aimed at achieving sustainable and decarbonized energy infrastructures. Since faults in PV panels can significantly impair system performance and reduce energy yield, early detection is essential to maintaining operational efficiency. This study proposes an advanced fault detection framework leveraging contemporary machine learning techniques to identify anomalies early, thereby preventing performance degradation and minimizing downtime. The proposed model incorporates single- and multi-objective feature selection strategies to enhance fault identification accuracy. Specifically, single-objective techniques utilize Information Gain (IG), Gini Ratio (GR), and Gini Importance (GI). At the same time, the Nondominated Sorting Genetic Algorithm II (NSGA-II) is employed as a multi-objective method to select an optimal subset of fault-relevant features. The model was trained on a dataset comprising 600 instances. Results indicate that gradient boosting achieved 97.6 % accuracy using single-objective selection, whereas the NSGA-II-based approach reduced the feature set to 15 optimal variables and improved accuracy to 99.3 %. Implementing this model enhances the operational reliability of solar PV systems and supports the broader goal of advancing the sustainability of solar energy production.