1.School of Automation, Wuhan University of Technology, Wuhan 430070, China;2.School of Electrical Engineering and Automation, Wuhan University, Wuhan 430072, China;3.State Grid Liaoning Electric Power Co., Ltd., Shenyang 110004, China

Improving the energy management capability of modern intelligent buildings is an important measure to promote energy saving and efficiency enhancement of power grid under the background of tight power balance. This paper proposes a data-driven energy saving control strategy for building heating, ventilation, and air conditioning (HVAC) systems, which avoids the dependence of traditional model predictive control on accurate thermodynamic modeling. First, on the basis of building thermal network simulation model, the input features are constructed by time-delay embedding method, and a multi-layer neural network model based on rectified linear unit (ReLU) is established to realize the multi-step prediction of indoor temperature time series. Then, for the time-of-use price of power grid, the rolling time-domain optimization model is constructed and reconstructed into the form of mixed-integer linear programming to realize the efficient solution of the optimization model in the finite control period. Finally, a building thermal simulation model is built based on Simscape to verify the effectiveness of the proposed control strategy. The simulation results show that the proposed strategy can meet the requirements of temperature comfort and improve the economic operation level of the building.