A modified neural network based predictive control for non-linear systems

Abstract

This paper proposes a modified neural network (NN) based predictive control strategy for non-linear systems. In this new control strategy, control signals are generated by minimising an objective function, which considers both the cumulative differences between a set-point and the output of the neural model and the control signal changes. By adding the control signal changes to the objective function, the output of the control system becomes smoother. As a result, more accurate control performances can be expected. Lastly, a typical non-linear system is introduced for simulation study. The standard back propagation (BP) algorithm is used to train the weights of the NN. By comparing the simulation results by different control strategies, the effectiveness of the proposed modified control strategy is fully illustrated.