Thermal and power generation performance of photovoltaic noise barriers with a heat sink using a deep neural network

Photovoltaic noise barriers (PVNBs) enable space-efficient solar energy utilization but are confined to elevating module temperature and reducing efficiency and lifetime. Despite its importance, intensive analysis on the direct correlation among irradiation, surface temperature, and power generation in PVNB has not yet been conducted. This study experimentally evaluated the thermal and electrical performance of three photovoltaic (PV) configurations with indoor and outdoor experiments: PV-only, PVNB, and PVNB with heat sink. Indoor experiments measured averaged surface temperatures of PV for various configurations, while outdoor experiments collected power-generation data which were used to train a deep neural network (DNN) that predicts power output using solar irradiation and averaged surface temperature of module. The PVNB exhibited the highest temperature rise, producing power reductions of 1.3 % at 700 W m⁻² and 0.8 % at 800 W m⁻² relative to PV-only, whereas PVNB with heat sink enhanced natural convection and recovered the barrier-induced losses, fully at 700 W m⁻² and by approximately 85 % at 800 W m⁻². The developed DNN model captured physical trend, achieving coefficients of determination of 0.8887 for training and 0.9217 for validation, and most predictions fell within a ±20 % error. Economic analysis reported that the heat sink retrofit of PVNB provided similar annual revenue but resulted in a 29.5 % higher net present value owing to the enhanced lifespan of the PV module caused by the reduction in averaged surface temperature. Despite the limitations of narrow climate conditions and using only single crystalline silicon module with one built-on PVNB design, this study provides essential insight and a practical tool for assessing PVNB retrofit and predicting thermal and electrical performances.