Multivariable adaptive control of unknown nonlinear dynamic systems using neural networks

Based on the approximation capability and generalization property of the backpropagation neural networks (BPNNs) for nonlinear mapping on a compact set, this paper presents a novel approach for designing adaptive neurocontroller (ANC) for unknown multivariable discrete nonlinear systems in general form. The key ideas of the proposed control strategy are of applying the Clarke's one-step-ahead weighted predictive control performance index and linearizing the feedforward BPNNs identifier models. Simulation results demonstrate the effectiveness of the new adaptive neural control scheme.<>