TWO NONLINEAR PREDICTIVE CONTROL ALGORITHMS BASED ON STRUCTURED NEURAL MODELS

This paper describes two nonlinear Model Predictive Control (MPC) algorithms with neural models of a special structure and their applications to a bioreactor. The model calculates the many-step ahead prediction without the necessity of being used recursively, the prediction error is not propagated. The first algorithm solves on-line a nonlinear optimisation problem. In order to reduce the computational burden, in the second algorithm the structured neural model is used on-line to determine the local linearisation and the nonlinear free trajectory. It is more computationally efficient and reliable, because only a quadratic optimisation problem has to be solved. Although being suboptimal, the second algorithm gives closed-loop control performance similar to that obtained in the first algorithm.