Experimental Data–Driven Machine Learning Prediction of Photovoltaic Panel Temperature

Abstract

Accurate prediction of photovoltaic (PV) panel surface temperature is essential for evaluating thermal behaviour and minimizing efficiency losses, particularly in hot and arid regions. This study presents an experimental and data-driven approach for predicting PV surface temperature using locally measured meteorological data from Nasiriyah City, southern Iraq. Field measurements were conducted over five consecutive days in early July under real outdoor conditions. The PV surface temperature was measured directly using thermocouples fixed on the front surface of the panels, while solar irradiance and ambient temperature were recorded using a solar irradiance meter and a data logger.

The experimentally measured dataset was used to develop and evaluate three regression-based machine learning models: Linear Regression, Random Forest, and Gradient Boosting. A time-based validation strategy was adopted to reflect realistic operating conditions, and model performance was assessed using the coefficient of determination (R²) and mean absolute error (MAE). The results showed that Linear Regression achieved the highest prediction accuracy, with an R² value of 0.9999 and an MAE of 0.028 °C, indicating a strong linear dependency between PV surface temperature and the selected meteorological variables. Although Random Forest and Gradient Boosting models also demonstrated good predictive capability, their higher error values suggest that increased model complexity did not improve performance for the investigated thermal regime.

The findings confirm that PV thermal behavior under steady outdoor conditions is predominantly governed by linear heat transfer mechanisms. Accordingly, simple and physics-consistent models can provide reliable and efficient temperature prediction for PV systems operating in hot-climate environments.