Research On Neural Network Based Adaptive Control Systems

Recently, most research on neural-network-based nonlinear control systems focus on the affine nonlinear system, which is really not as general as nonaffine nonlinear system in practical production. So neural network identification and neural-network-based adaptive control for a class of nonaffine nonlinear system will be researched in this paper.First this paper gives an over view of the recent studies on neural network-based adaptive control systems. And several important problems existing in NN adaptive control systems are also proposed.Based on the analysis of neural network identification models of a class of nonlinear system and the discussion of modified BP algorithms and recursive least squares (RLS) algorithm, a kind of inner recurrent state identification model and its neural network training algorithms are proposed. Its identification performance and comparison between algorithms are demonstrated in some simulation results.Next a novel adaptive reference model tracking control system design approach for the nonlinear system is discussed and then the existing conditionsof the tracking control scheme is presented. Based on that, the design of feed forward control matrix and the iterative regulative algorithms of feedback control matrix are put forward, which may lead the system to meet a good tracking performance. Finally, neural-network-based adaptive reference model tracking control for unknown systems is deeply discussed. Simulation studies of electrode regulator systems of arc furnace show the effectiveness of the proposed control approach.