Intelligent fault diagnosis and simulation-based modelling of grid-connected photovoltaic systems under desert dust conditions

Abstract

Photovoltaic (PV) power plants operating in desert environments experience continuous efficiency losses because dust accumulation gradually reduces solar radiation reaching the module surface, leading to lower energy production even under favourable weather conditions. Accurate prediction of normal operating behaviour therefore provides a reference for distinguishing genuine faults from natural fluctuations in plant performance. This study proposes a data-driven framework for PV power prediction and residual-based fault diagnosis at the Aoulef PV power plant in southern Algeria. Artificial neural networks (ANN) and adaptive neuro-fuzzy inference systems (ANFIS) were developed from measured solar irradiance and ambient temperature to estimate the healthy-state PV power output. Model performance was assessed through regression analysis and statistical error indices, whereas fault detection relied on residuals computed from the difference between measured and predicted power. Both models achieved excellent prediction accuracy, with coefficients of determination approaching 0.99. The ANN produced lower prediction errors than the ANFIS, with a root mean square error of 0.009 and an mean absolute error of 0.004, which improved the sensitivity of the residual-based diagnosis. Dust accumulation, identified as fault F13, generated clear residual deviations that enabled automatic fault detection without interrupting plant operation. The findings indicate that the ANN framework combines high predictive accuracy with low computational demand, offering a practical and reliable solution for intelligent monitoring and maintenance of PV systems operating under harsh desert conditions.