The Synthesis And Application Of Adaptive Neural Network In Iterative Learning Control And Decentralized Control

In this paper, based on adaptive neural network control theory, the adaptivelearning control with adaptive nonlinearly parametric variable wavelet neural network,adaptive decentralized neural control for a class of pure-feedback large-scaleinterconnected systems and a class of large-scale interconnected systems withtime-varying delays and input delays are researched, respectively. The maincontributions are summarized as follows.Firstly, the problem of adaptive learning control for a class of unknown nonlinearsystems is addressed on the base of nonlinearly parametric wavelet neural network.Anovel adaptive learning control strategy is proposed. A robust adaptive nonlinearlyparametric neural network controller is designed. The unknown nonlinear functions areapproximated by variable wavelet neural network, in which the number of nodesgradually increasing following the iteration times. Based on several modifiedassumptions on the nonlinearly parametric variable neural network, the difficulties,designing the adaptive algorithm for nonlinear parameters in variable neural networkunder the variation of structure of neural network is dealt with by using Lyapunovmethod. The convergence of tracking error with iteration is proved. Moreover, thefiniteness of all the signals is guaranteed. Finally, the effectiveness of the proposedmethod is illustrated with a simulation example.Secondly, for a class of pure-feedback interconnected system with unknowntime-varying delays in outputs interconnections, an adaptive decentralized neuralcontroller is designed. By taking advantage of implicit function theorem and themean-value theorem, the difficulty from the pure-feedback form is overcome. Under awild assumption that the nonlinear interconnections are assumed to be bounded byunknown nonlinear functions with outputs, the difficulties from unknowninterconnections are dealt with, by introducing hyperbolic tangent functions. Toeliminate “the explosion of complexity” problem in backstepping procedure, adynamic surface control technique is applied. In addition, minimal learning parameterstechnique is successfully incorporated into this novel control design to reduce thecomputational burden. With an appropriate Lyapunov-Krasovskii functional, thesemi-global uniform ultimate boundedness of all the signals in the closed-loop systemis guaranteed. Finally, simulation studies are given to demonstrate the effectiveness of the proposed design scheme.At last, for the input delays problem in large-scale interconnected system, anadaptive decentralized neural network controller is designed. Compared with thetraditional backstepping design procedure, dynamic surface control technique isintroduced to eliminate “the explosion of complexity” problem. To reduce thecomputational burden, minmal learning parameters technique is developed, also. Bydesigning an appropriate input delays compensated controller, the input delays problemis addressed. Under a novel Lyapunov-Krasovskii functional, the semi-global uniformultimate boundedness of all the signals in the closed-loop system is guaranteed. Finally,a simulation example is given to illustrate the effectiveness of this design method.