Nonlinear system predictive control using a multi-dimensional Taylor network

Abstract

A predictive control scheme on the basis of multi-dimensional Taylor network (MTN), named as MTN predictive compensation control, is proposed for single-input single-output nonlinear systems in this paper. We consider the MTN model as a one-step-ahead predictive model and train it by back-propagation (BP) algorithm with a momentum term, and then control the system by the predictive control law. Furthermore, to improve the anti-disturbance performance of the system, another MTN model is considered as the compensator and trained by a recursive least-squares algorithm to counteract the disturbance. An example is used to verify the effectiveness of the proposed scheme, in which the noise disturbance is considered. For comparison, the scheme that combines a radial basis function (RBF) neural network with a proportional-integral-derivative (PID) controller (RBF-PID), the predictive control scheme that combines a predictive neural network with a control neural network based on a BP neural network, and MTN predictive control are introduced. In addition, the difference in computational complexity between the MTN predictive compensation control scheme and the RBF-PID scheme is also considered. The experimental results show that the proposed scheme is effective, has good anti-disturbance performance in predictive control for the nonlinear system and is superior to the other comparison schemes.