Nonparametric Identification and Predictive Neurocontrol of Uncertain Systems

The aim of this paper is to present the main features of an integrated approach of the on-line identification and predictive control of uncertain dynamical systems. A nonparametric statistical procedure is considered for the functional estimation of the unknown process model, imbedded into a neuralbased predictive control procedure: a feedforward neural network with one hidden layer is used to generate an optimal receding horizon control strategy, by minimizing a constrained predictive-control-like criterion. The topology of the network controller is beforehand optimally selected off-line by a Bayesian statistical procedure. Applications to simulated case studies in the field of bioprocess control are proposed.