Chapter 5 - Intelligent Methodology

Abstract

Chapter 5, first, studies a generalized procedure in the identification and control of a class of time-varying, delayed, nonlinear dynamic systems. Under the framework, recurrent neural network is developed to accommodate the online identification, which the weights of the neural network are iteratively and adaptively updated through the model errors. Then, indirect adaptive controller is designed based on the adaptive law of the controller, which the adaptive law of the controller is designed for the controller design. Second, it studies a neural network–based finite-time adaptive controller for a class of uncertain nonaffine nonlinear systems in pure feedback prototype. In this framework, homeomorphism mapping and finite time convergence are introduced to ensure the state constraint and convergence in finite time, respectively. To prevent constraint violation, the homeomorphism mapping is applied to a chosen compact superset, which contains initial condition and virtual superset without constraint, then use the finite time converge method to ensure the converge time after the mapping transform. Accordingly, it provides a generic prototype for finite time adaptive law to train neural networks.