Fault detection and diagnosis in photovoltaic power systems using fisher random matrix approach

Abstract

This paper proposes a novel methodology for fault detection and diagnosis (FDD) in photovoltaic (PV) systems that combines Fisher's linear discriminant (FLD) with the Mahalanobis distance. The approach utilizes FLD to reduce the dimensionality of operational data while maximising the separation between healthy and faulty states. The Mahalanobis distance is then used to detect anomalies by accounting for correlations among variables such as voltage, current, and temperature. The validity of the method was established using real-world data from a 20 MWp PV plant in Algeria. The results obtained demonstrate the efficacy of the proposed method in classifying various faults, including open-circuit, shading, and short-circuit faults. The approach demonstrates substantial improvements in detection accuracy, efficiency, and false-alarm reduction compared to conventional methodologies. The proposed FDD solution is robust and scalable, rendering it ideal for real-time monitoring of large-scale PV systems.