Dual version of neural adaptive controller for stochastic nonlinear systems with functional uncertainty

A dual adaptive control scheme for general class of nonlinear stochastic discrete control systems is proposed. It enhances the controller's capacity for adaptation, stabilization, management of uncertainty and learning in more complex and uncertain plants. The control law in the proposed dual adaptive control scheme is derived by taking into consideration model uncertainty. This is achieved by extending techniques from the stochastic adaptive control, known to be computationally expensive, to the functional adaptive case leading to a novel dual control laws for neural control schemes. The proposed scheme avoids the pre-control neural network training phase by taking into consideration model uncertainty and its effect on tracking, in the on-line control phase. Its implementation requires finding a suitable method for estimating model uncertainty. Both simulation and experiment are given in the paper to demonstrate the efficiency of the proposed method.