Research on residual heat prediction and energy management control of photovoltaic greenhouse based on data-driven method

Ventilation-induced residual heat loss is a key factor affecting the energy utilization efficiency of greenhouses. The residual heat in photovoltaic greenhouses is influenced by multiple meteorological factors, such as temperature, humidity, wind speed, and solar radiation, which exhibit complex nonlinear relationships. Although traditional physics-based models perform well under specific conditions, they exhibit significant limitations in handling the nonlinear relationships among multiple parameters. With the rapid advancement of artificial intelligence, data-driven approaches have increasingly become effective tools for addressing such nonlinear problems. To address this challenge, this study proposes an innovative system that combines a PV-driven heat pump with waste heat recovery, integrated into a 24.5 m² PV greenhouse. A dynamic model between solar radiation, crops, and the heat recovery system was established to determine the greenhouse’s temperature and estimate the actual heating and cooling demands of the crops. In addition, data-driven approaches, including MLP, CNN, and ResNet models, were used to predict the residual heat energy in the greenhouse at different time scales, and the energy migration characteristics of the heat recovery system under different working conditions were analyzed. The results indicate that as the time scale increases, the ResNet model consistently demonstrates superior predictive performance across different time scales, achieving R² values ranging from 0.890 to 0.929. The proposed system improves the primary energy ratio by 27.0 % and the system coefficient of performance by 36.9 %, with an overall energy saving rate of 79 %. This study provides valuable insights for residual heat prediction and energy management in photovoltaic greenhouses.