heating, ventilation, air conditioning systems (HVACs), as one of the load types with the largest proportion of electricity consumption in urban buildings, have great adjustment potential and can participate in demand response to promote the balance of supply and demand in the power system. Based on the existing modeling foundation of HVACs, this paper analyzes the physical model structure of each subsystem, accurately selects the appropriate features through the physical model structure, and provides guidance for the data-driven method. Combined with the MLP network stucture, a knowledge-driven and data-driven approach is proposed to modelHVACs. Then, based on the established HVACs model, the HVACs energy consumption optimization model is established. And MLP network structure is made explicit, so the HVACs energy consumption optimization model is transformed into a mixed integer linear programming problem by linearizing the activation function. Cases analysis show the it can be seen that the modeling method proposed in this paper can quickly and accurately select features, which is beneficial to improve the modeling efficiency. In addition, this control method greatly reduces the difficulty of solving the control model, and the practical application higher value.