Explainable fault diagnosis using discrete grey wolf optimization algorithm for photovoltaic system

Abstract

The present article introduces the discrete grey wolf optimization algorithm (DGWOA), a novel variant of the grey wolf optimizer (GWO). DGWOA integrates discrete optimization techniques with explainable artificial intelligence (XAI) methodologies. This approach aims to overcome limitations associated with traditional fault diagnosis methods, such as limited accuracy in identifying complex patterns and low interpretability. Furthermore, it mitigates early convergence problems commonly encountered in optimization algorithms and enhances adaptability to discrete classification challenges. The DGWOA algorithm is designed to generate interpretable classification rules for fault detection through a stochastic search strategy. The explainability provided by the model not only enhances decision-making transparency but also improves diagnostic efficiency and predictive accuracy. The proposed algorithm was evaluated using a photovoltaic system dataset and benchmarked against established rule-based classifiers. DGWOA consistently achieved a classification accuracy of 99.48% and a precision of 100%, demonstrating its effectiveness in enhancing fault detection. Moreover, the interpretability of the generated classification rules contributes to the generation of outcomes that are both actionable and comprehensible to decision-makers.