In view of the characteristics of variable air volume (VAV) air conditioning system which is multiple variable, nonlinear, large delay and time-varying. In the control of air conditioning if we only use one kind of traditional intelligent control method to deal with the factors such as disturbance and uncertainty can not get good control effect. In this paper, the multiple mode control method is introduced in VAV air conditioning system. This method sets up two kinds of control modes, chooses different controlled objects under different disturbance in order to achieve rapid adjustment and characteristic of energy saving. It combines fuzzy neural network and predictive control to establish a neural network prediction model, and uses the difference between the predictive output and the actual output to adjust the parameters of the predictor online. The results show that the control effect of the multiple mode control system is better than the traditional neural network predictive control system. Multiple mode control method has the characteristics of strong robustness, high control precision, strong adaptive ability, safe, reliable and energy saving. The system has wide application prospect.