An Effective Model Predictive Control with Self-Optimizing Deep Belief Network

Model predictive control (MPC) is an effective control based on the state prediction and rolling optimization. However, the system states are from the predictive model, and these existing predictive models fail to give the accurate modeling and prediction results of system states. In this paper, we propose an effective model predictive control with self-optimizing deep belief network (MPC-SODBN). First, a self-optimizing deep belief network (SODBN-EL) is designed based on the event-triggered learning to improve the performance of modeling and predicting system states. Second, we design a loss function that consists of the output of SODBN-EL and the control law. The optimal control law sequence can be obtained based on the loss function. Finally, the simulation experiments on the nonlinear system modeling and control show that the MPC-SODBN is effective and efficient.