Air conditioning energy consumption plays an important role in the whole building energy consumption. The establishment of air conditioning system energy consumption model can help building operators make accurate energy management plans, optimize the operation of building energy consumption equipment, and realize intelligent energy management. In this paper, a combined prediction model for energy consumption of hybrid air conditioning system is proposed. Particle Swarm Optimization (PSO) is used to optimize the super parameters of Multi-Layer Perceptron (MLP). Boruta was used to select the features of the data set, and PSO-MLP model was established to predict the energy consumption of the system. The results show that the energy consumption prediction model based on Boruta+PSO-MLP can more accurately predict the energy consumption of building air conditioning. Compared with the single MLP model, Mean Absolute Error and Root Mean Squared Error decrease by 57.3% and 49.7%, respectively, and the correlation coefficient reaches 0.962.