CFD-Based Machine Learning Model for Agrivoltaic System Design

Agrivoltaics, the co-location of solar and food production, is a promising solution to land-use conflict between solar photovoltaics (PV) and agriculture. Microclimate studies indicate that agrivoltaic systems enhance solar farm cooling, leading to increased panel efficiency, but stakeholders lack efficient design tools to quickly evaluate the consequences of various agrivoltaic designs. Here we present a computational fluid dynamics (CFD)-based machine learning (ML) model which is utilized to develop an early version of an agrivoltaic design tool, where solar cell operating temperature is optimized based on panel height and ground cover type selection. Results indicate that Random Forest Regression (RFR) and Gradient Boosted Trees (GBT) perform better than Support Vector Regression (SVR) and Linear Regression (LR) in this case, with RFR and GBT achieving under 2 °C RMSE compared to SVR and LR over 3 °C. Using dual annealing optimization with the GBT model, findings indicate that panel heights up to 3.2m and ground albedo up to 64% are preferred in warm months to maximize panel cooling.