Water-cooled VRF system is a strongly coupled, highly nonlinear complex system. The main parameters of the refrigeration system components coupled, interrelated and closely associated with the indoor environment. Using classical control theory can't be well to improve the performance of water-cooling VRF. In this article, neural network theory is used to study the control strategy, and the results show that the control strategy can be applied to dynamic simulation of water-cooling VRF. It provides the basis for the development of water-cooling VRF.