Performance assessment and dynamic fault detection in photovoltaic systems using artificial intelligence

The intermittent nature of photovoltaic (PV) solar energy, primarily due to variable weather conditions, results in significant power losses (averaging a 25 % decrease in energy production) and system failures, underscoring the need for reliable performance assessment. Precise characterization of losses and effective fault detection are crucial for informed decision-making in PV system optimization. This work introduces a computational model for evaluating PV system performance, which integrates a built-in fault detection mechanism. The model employs a dynamic loss quantification algorithm to assess meteorological, operational, and technical data. An artificial neural network (ANN) estimates the expected electrical production, where data points are derated by identified losses and faults if the ANN flags a faulty condition. Following this, three dynamic non-faulty operational threshold alternatives classify data points as normal or faulty. The efficacy of this model was validated using a case study of the PV system at Universidad de los Andes in Colombia. The primary contributions of this research include: a computational model demonstrating a 3.3 % root mean square error in AC power estimation and a 0.6 % mean absolute percentage error in daily energy estimation, an artificial intelligence algorithm for estimating expected electrical production with a median absolute percentage error below 5 %, and a fault detection mechanism achieving a 93.8 % accuracy and a 98.8 % recall.