Machine learning-based approach for predicting the PV modules temperature: A multi-technological assessment including soiling impact, toward a better solar plants’ operation under desert conditions

Operating temperature is a critical parameter influencing the efficiency and durability of photovoltaic (PV) systems, particularly in desert and semi-arid regions where intense solar irradiance, elevated ambient temperatures, and frequent soiling prevail. Excessive module heating not only reduces electrical conversion efficiency but also accelerates material degradation, making accurate temperature prediction essential for improving system performance, reliability, and lifespan. This study contributes to the advancement of efficient PV deployment in harsh climates by developing a machine learning (ML) model capable of accurately predicting PV module temperature under real outdoor conditions. The model is based on the Extreme Gradient Boosting (XGBoost) algorithm and is trained on a comprehensive, high-resolution dataset collected over one year in the hot semi-arid climate of Benguerir, Morocco. A key novelty of this work lies in its multi-technology and multi-condition modeling approach: it simultaneously predicts the operating temperature of two widely deployed PV technologies, polycrystalline silicon (pc-Si) and cadmium telluride (CdTe), while explicitly accounting for the impact of natural soiling, using both clean and soiled modules from each technology. For benchmarking, a multiple linear regression (MLR) model was developed using the same input features. Results show that the XGBoost model achieves high predictive accuracy across all configurations, with a coefficient of determination (R2) of 0.9869, significantly outperforming the MLR model (R2 = 0.8963). Seasonal and weather-specific evaluations further confirm the robustness of XGBoost, with relative deviations consistently within ±5 % for all module types and conditions. In contrast, the MLR model exhibits substantial errors, particularly during clear-sky periods in the wet season, where deviations exceeded −30 %. Year-long daily comparisons also reveal that XGBoost maintains stable performance across technologies, seasons, and soiling levels, highlighting its effectiveness as a predictive tool for PV thermal behavior in harsh climates. These findings underscore the potential of advanced AI-based modeling as a powerful and reliable tool for predicting PV thermal performance, aiding in better system design, performance optimization, and thermal management in challenging desert environments.