Optimal control for nonlinear systems by a neural controller of the state feedback type

This paper is concerned with designing a neuro-controller for optimal control of nonlinear systems. The neuro-controller approximates the optimal state feedback control law for the nonlinear systems. The neuro-controller consists of a multilayer neural network and its matrices of connection weights are optimized so that the performance function is minimized. Optimality conditions for the neuro-controller is derived first and then optimal values of the matrices of connection weights are calculated by a steepest descent method with error backpropagation algorithm. Learning of the weights are carried out by off-line scheme.