Modeling Of Nonlinear Time-Varying Systems By Using Time-Varying Neural Networks

The nonlinear time-varying systems exist widely in actual life, for nonlinear time-invariant systems, many identification methods which are effective have been developed. But, when the system parameters are time-varying, the extant identification methods only can track the time-varying parameters, and achieve bounded convergence. It is a challenging research topic to consider the method of nonlinear time-varying systems modeling.In recent years, different kinds of neural networks and study algorithms are developed and comleted, nonlinear time-varying systems modeling methods based on neural networks receive more attention from scholars, and lots of methods are presented. In this paper, we consider the modeling of nonlinear time-varying systems methods by using time-varying neural networks. Take into account the issues above mentioned, this thesis focuses on the following aspects:1. Considering nonlinear time-varying systems modeling by using time-invariant neural networks, several improved algorithms of BP algorithm are introduced. Focus on RLS algorithm of RBF networks and L-M algorithm of BP neiworks, derive its implementation process. At last, simulations are presented to demonstrate the feasibility.2. Considering nonlinear time-varying systems modeling methods, based on RBF neural networks, the concept of time-varying weights is presented, we get the modeling method based on time-varying RBF neural networks. Iterative learning idea is introduced into the training of time-varying RBF neural networks (TVRBFNN). The iterative learning least squares (ILLS) algorithm is derived for updating time-varying weights, and the theoretic properties of ILLS algorithm are discussed. Modeling nonlinear time-varying systems by using time-varying RBF neural networks, and compare with the simulation results of modeling nonlinear time-varying systems by using time-invariant RBF neural networks and time-varying RBF neural networks, we can know that the latter method is better.3. Introduce the ILLS algorithm into other neural networks structure, such as Chebyshev, Legendre, Hermite and Polynomial neural networks, get their time-varying basis function neural networks. In the paper, apply ILLS slgorithm in to single-input time-varying Chebyshev, Legendre, Hermite neural networks and two-input time-varying Polynomial neural networks, numerical simulation results are presented to demonstrate the effectiveness of ILLS algorithm in these neural networks.