Maintenance Decision Modeling for Solar Photovoltaic Performance Prediction Using RMSE and MAPE Methods

This study evaluates the efficiency and degradation of a 999.50 kWp photovoltaic (PV) system in Thailand under clear sky conditions. It found that the system efficiency ratio significantly decreased, 10% below the design threshold. Thus, a feasible and optimal model can be identified to support better maintenance decisions. In addition, a real-time performance monitoring approach is proposed to enable building operators to make informed decisions about mechanical maintenance or cleaning maintenance. This method involves comparing the actual power generation data with the efficiency prediction based on linear regression modeling. In addition, the impact of dirt accumulation on the efficiency of the PV modules is evaluated. The preliminary results indicate that the output power decreases to 213.74 kW, and the corresponding efficiency ratio is 46.76%. A linear regression-based warning mechanism is adopted to issue an alarm when the efficiency drops below a threshold of 50% ± 10 or when the power drops below 300 kW ± 50. The results show that using this prediction approach can improve the system efficiency ratio by up to 70% and increase the output power by more than 355.53 kW. The accuracy of the model is verified with a mean square error(RMSE) of 18.86% or more than 81.14% and a mean absolute percentage error(MAPE) of 4.56%. The study concludes that the prediction simulation can effectively guide maintenance interventions, which can improve both power generation and system efficiency. Future work may incorporate AI-driven predictive analysis to further optimize the PV system and maintenance strategies.