ADAPTIVE MODEL-BASED CONTROL USING NEURAL NETWORKS

Identification and control of nonlinear processes can be achieved usin g neural networks. However, previous work extending the identification to the on-line adaptive case has resulted in extremely poor performan ce, preventing practical application in a control framework. The work presented in this paper proposes and demonstrates a powerful method fo r implementing a neural network model of nonlinear process dynamics fo r adaptive control. The performance of adaptation has now been suffici ently raised to allow practical adaptive control to be considered. The new adaptive method has been amalgamated with multistep nonlinear pre dictive control techniques to form an adaptive neural controller. The performance of this controller is demonstrated, and evaluated using tw o simulated realistic processes; level control of a conical tank and m ultivariable control of an industrial evaporator. The results indicate that these techniques have good practical potential for the adaptive control of nonlinear processes.