Abstract-In the predictive control, it is often need to obtain the model which prediction accuracy is possible higher, however, in practice controlled objects are often exist nonlinear, parameter variation, model mismatch, disturbance and other factors, there is a big error between the controlled object-based prediction model and the actual output of the object. In this paper, using neural network to improve prediction model accuracy, at the same time feedback correction method is used to compensate for the lack of neural network prediction model, simulation results show that this method has better accuracy than conventional feedback correction.