Forecast-based predictive analytics and optimized maintenance digital twin for utility-scale photovoltaic power plants

To ensure optimal lifetime performance and high-performance operations for photovoltaic (PV) systems, it is essential to anticipate underperforming conditions and implement cost-effective maintenance strategies. This study presents a novel forecast-based predictive analytics and maintenance digital twin (DT), addressing the challenge of accurately performing predictive methods for potential faults and loss conditions in utility-scale PV power plants. The proposed DT model leverages high-resolution data streams, operates both machine learning (ML) and physical models in parallel and evaluates deviations in electrical parameters and performance trends through systematic analysis of anomalies and rule-based logic. The results indicate that the developed predictive model achieved power output prediction errors of less than 2%, as determined by the average root mean square error (RMSE) relative to the nominal capacity. This level of accuracy renders the model highly suitable for real-time predictive analytics applications. At the utility scale, the model demonstrated high reliability in predicting long-term performance losses for insulation resistance (R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iso</inf>), potential induced degradation (PID), and soiling. Specifically, the forecast-based trend evaluation algorithm proved capable of predicting PID loss decreases once the difference between the predicted and observed daily average voltage slopes deviated >2%. Similarly, the predictive model for monitoring R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iso</inf> demonstrated effectiveness in forecasting low insulation incidences by analyzing the daily R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iso</inf> patterns. Finally, the findings offer valuable insights into advanced performance monitoring, emphasizing the potential of integrated predictive analytics for utility-scale PV plant monitoring aligned with smart grid paradigms.