Dynamic matrix control based on neural networks error compensation

Predictive control of multi-step prediction and rolling optimization have the ability to overcome the impact of modeling errors, giving the control system a certain robustness. However studies show that if the model is a mismatch, the robustness of predictive control is limited and we still need a more accurate prediction model. Prediction error compensation based on the principle of feedback correction is an effective method to improve the accuracy of the prediction in order to enhance the robustness of a control system. In this paper, a kind of dynamic BP network with online adjusted weight value is used to fit the predicted model error. Together with the predictive model, this constitutes a dynamic prediction combination. Eventually, a predictive control algorithm with dynamic compensation capacity is acquired. The algorithm significantly improves the prediction accuracy, increasing the robustness of the predictive control algorithm.