Intelligent predictive control of nonlinear processes using neural networks

Abstract

This paper presents a novel approach to design of generalized predictive controllers (GPC) for nonlinear processes. A neural network is used for modelling the process and a gain-scheduling type of GPC is subsequently designed. The combination of neural network models and predictive control has frequently been discussed in the neural network community. This paper proposes an approximate scheme, the approximate predictive control (APC), which facilitates the implementation and gives a substantial reduction in the required amount of computations. The method is based on a technique for extracting linear models from a nonlinear neural network and using them in designing the control system. The performance of the controller is demonstrated in a simulation study of a pneumatic servo system.