Hybrid fault detection and identification strategy for PV systems combining statistical data-driven and Kalman filter algorithms

• Hybrid FDD for Real-Time PV Monitoring: Combines regression and Kalman Filter for robust fault detection. • Two-step data-driven robust extraction of diagnostic indicators from I-V curves. • MZP Adaptive Statistical Thresholding capturing nonlinear effects of temperature for Robust Fault Detection. • Boolean-based method for precise fault identification: robustness and interpretability. This paper presents a novel hybrid fault detection and diagnosis (FDD) methodology for real-time monitoring of the DC side of photovoltaic (PV) systems. The approach combines data-driven statistical regression with a Kalman Filter (KF) to detect early and small-scale faults, even under low irradiation, while accounting for model uncertainties and measurement noise. Three diagnostic indicators derived from I-V curves are estimated using a two-step regression method that corrects for temperature effects and predicts indicators across all irradiation levels. A statistical adaptive threshold, derived from regression uncertainties, enhances sensitivity to soft faults and minimizes false alarms to detect short-circuit, open-circuit, and incipient series and shunt degradations. A novel classification algorithm further identifies faults through a fault signature matrix based on threshold violations. Validation via MATLAB/Simulink® simulations under diverse weather conditions demonstrates a short mean detection delay of 0.21 s and an overall classification accuracy of 90.16 %, highlighting the method’s robustness, adaptability, and effectiveness across a wide irradiation spectrum.