Interpretable Rule-based Classification for Photovoltaic Fault Detection Using Discrete Tuna Swarm Optimization

Abstract

The reliability of photovoltaic systems is dependent on effective fault detection and diagnosis, which are essential to maintain performance, efficiency, and operational safety. Traditional diagnosis methods can achieve good accuracy but often lack interpretability, which limits their use in real-time monitoring and decision-making. In this study, we propose a framework called Rule-based Classification Fault Detection and Diagnosis for Photovoltaic Systems (RCFDD-PVs). The framework incorporates a rule-based classification approach in which IF–THEN decision rules are intelligently generated using Discrete Tuna Swarm Optimization. This hybrid white-box approach ensures that diagnosis decisions are not only accurate but also understandable, offering a clear alternative to black box models. The framework was validated on a dataset including both normal and faulty PV operating states. Results show that RCFDD-PVs achieved 99% accuracy and a weighted F1-score of 0.99, using only nine rules with four antecedents to achieve complete dataset coverage. The generated rules are concise and interpretable, making them practical for integration into PV monitoring systems.

Keywords