Fault Detection and Classification Methodology for Photovoltaic Systems Through AI-based Digital Twin

This paper presents a Fault Detection and Classification (FDC) methodology using an Artificial Intelligence-based (AI-based) Digital Twin (DT). Initially, the AI-based DT simulates the generated photovoltaic (PV) power under specific environmental conditions. Subsequently, the FDC methodology assesses the system state and classifies faults. The case study in the paper considers four types of faults (shading, open circuit in one string, open circuit in two strings, and open circuit in three strings). Additionally, the time series of PV power production used in this study were simulated using MATLAB Simulink, considering real meteorological conditions with a recording frequency of one hour. The simplicity of the proposed methodology ensures its applicability to large-scale PV systems, with results indicating up to 93.6% fault classification accuracy and 100% fault detection accuracy.