Stochastic learning control for nonlinear systems

In this paper a new learning control algorithm for complex systems are proposed. This control algorithm is based on i) artificial neural network; ii) a quadratic control criterion. The neural network plays the role of the controller and their weights are adjusted using stochastic approximation techniques. Both unconstrained and constrained control objectives are considered. The Lagrange approach is used to deal with the constrained case problem. This control strategy presents another characteristic: robustness. It is able to deal with process parameters variation. No process model is used for control purposes. The feasibility and the performance of the control algorithm are illustrated by an example: the control level of a conic tank that exhibits a high nonlinearity characteristic.