Detection and Clustering of Photovoltaic Faults

Photovoltaic (PV) systems are susceptible to faults that diminish energy yield and efficiency. This paper presents a novel framework for detecting and clustering PV faults using accessible data: maximum power point voltage and current from the inverter's DC side, plus weather time series. Fault detection uses a degradation-adaptive model to estimate expected outputs, flagging anomalies via dynamic irradiance-based thresholds and performance filters. Fault impacts are quantified with current and voltage performance indices (CPI and VPI), serving as clustering features. Among tested algorithms (K-means, FCM, DBSCAN, HDBSCAN), HDBSCAN excels in robustness to noise on simulated data, grouping faults by system effects.