Neural network based multi-step predictive control for nonlinear systems

A neural network model-based multi-step predictive control algorithm with its convergence analysis for nonlinear discrete systems is presented. The structure of this nonlinear system is separated into a simple linear part and a nonlinear part by using the local linearization of nonlinear activation functions. The proposed algorithm gives a direct and effective multi-step predicting method. It uses simple linear predictive control methods to get the control law and avoids the complicated nonlinear optimization. Simulation results show the efficiency of this method.