Learning Model Predictive Control Law for Nonlinear Systems

In this paper, the model predictive control(MPC) problem of nonlinear systems is studied. Since the traditional MPC algorithms require a lot of computing resources and time to solve optimal control problems, it is hard to use in practical systems with high sampling rate. To resolve this issue, this paper proposes a deep neural network-based learning model predictive control (LMPC) algorithm to improve the speed of computing the control laws. Because the use of deep neural network to learn the control law instead of solving optimization problems, the computation speed of MPC algorithm has been greatly improved. Simulation experiments verify the effectiveness of the proposed algorithm.