A multi-scale smart fault diagnosis model based on waveform length and autoregressive analysis for PV system maintenance strategies

Nonlinear photovoltaic (PV) output is greatly affected by the nonuniform distribution of daily irradiance, preventing conventional protection devices from reliably detecting faults. Smart fault diagnosis and good maintenance systems are essential for optimizing the overall productivity of a PV system and improving its life cycle. Hence, a multiscale smart fault diagnosis model for improved PV system maintenance strategies is proposed. This study focuses on diagnosing permanent faults (open-circuit faults, ground faults, and line-line faults) and temporary faults (partial shading) in PV arrays, using the random forest algorithm to conduct time-series analysis of waveform length and autoregression (RF-WLAR) as the main features, with 10-fold cross-validation using Matlab/Simulink. The actual irradiance data at 5.86 °N and 102.03 °E were used as inputs to produce simulated data that closely matched the on-site PV output data. Fault data from the maintenance database of a 2 MW PV power plant in Pasir Mas Kelantan, Malaysia, were used for field testing to verify the developed model. The RF-WLAR model achieved an average fault-type classification accuracy of 98 %, with 100% accuracy in classifying partial shading and line-line faults