The temperature difference between day and night in a solar greenhouse is large. Heat in a greenhouse is typically in excess during the day while the temperature is low and the humidity is high at night. This study designs and tests an active heat storage and release air-source heat-pump system with a thermally insulated water tank as the energy storage body. By comparing air temperature and humidity in a test greenhouse with a control greenhouse in typical weather conditions, the power consumption and performance of the system are evaluated. The results show that compared with the control greenhouse, the average daytime temperature of the test greenhouse is lowered by about 3 °C during the operation of the system in typical weather conditions. At night, the average temperature is increased by about 4 °C, and the relative humidity is decreased by about 20%. When optimized, the maximum coefficient of performance (COP) of the system can reach 4.32 in heat storage mode. The nighttime heat release from the energy storage tank accounts for 26.9% to 51.2% of the nighttime energy consumption, and the energy utilization efficiency is 59.6% to 497.0%. This study provides a new way to control environmental parameters in solar greenhouses.

The high energy consumption CEA building brings challenges to the management of the energy system. An accurate energy consumption prediction model is necessary. Although there are various prediction methods, the prediction method for the particularity of CEA buildings is still a gap. This study proposes some derived features based on the CEA scenarios to improve the accuracy of the model. The study mainly extracts the time series and logical features from the agricultural calendar, the botanical physiological state, building characteristics, and production management. The time series and logical features have the highest increase of 2.8% and 3.6%, respectively. In addition, four automatic feature construction methods are also used to achieve varying degrees of influence from −9% to 8%. Therefore, the multiple feature extraction and feature construction methods proposed in this paper can effectively improve the model performance.