Modeling and Optimization of Solar Panel Cooling using Machine Learning and Deep Learning Techniques

Abstract

    Photovoltaic (PV) modules experience a marked efficiency drop when the cell temperature exceeds the nominal operating range. This study proposes an AI-driven cooling control framework that couples short-term temperature forecasting with adaptive decision-making to maximize the net energy yield under tropical operating conditions. This study used a six-month, 1-minute-resolution dataset collected near Kuala Lumpur, Malaysia, integrating IoT field measurements (PV rear-surface temperature, on-site irradiance, ambient temperature, relative humidity, wind speed, and inverter energy output) with meteorological/irradiance archives (MetMalaysia and the NREL Solar Radiation Database), resulting in approximately 2.6×10^5 time-stamped samples. After preprocessing (1.8% missing data handled via interpolation and KNN imputation; z-score standardization), a convolutional neural network (CNN) predicted the near-future PV surface temperature, and a reinforcement learning (RL) agent selected the cooling mode (passive, low-power fan, high-power fan, or water spraying) to balance thermal mitigation against auxiliary energy consumption. CNN achieved an RMSE, MAE, and R² of 1.38°C, 1.04°C, and 0.95, respectively, enabling anticipatory control that reduced unnecessary switching.