Dynamic Neural Network Identification And Predictive Control For Nonlinear Time-delay System

Numerous processes in modern industry are characterized by strong nonlinearity and such as high purity distillation columns modeling,high exothermic chemical reaction process,pH neutralization process,cooling treatment,biochemical process and so on.Moreover,the problem of identification and control of nonlinear time-delay systems has been widely concerned by researchers.Based on previous research,this paper uses dynamic neural networks to deal with nonlinear time-delay systems.The main research content of this paper is as follows:1.In this paper,the dynamic neural network namely dynamic extreme learning machine and recurrent neural network are used to identify nonlinear time delay systems,and a dynamic extreme learning machine algorithm and a quasi-linear recurrent neural network algorithm are proposed,this paper focuses on the quasi-linear recurrent neural network algorithm.The innovation of the dynamic extreme learning machine presented in this paper is that it only has two input nodes in its structure,and then it introduces adjustable time-delay parameters between its input nodes and hidden layer nodes,and the time-delay parameters are optimized by the particle swarm optimization algorithm.It is particularly suitable for dealing with nonlinear systems with time delays.But,due to the disadvantages of dynamic extreme learning machine such as overfitting and strong randomness,this paper proposes a quasi-linear recurrent neural network method based on recurrent neural network.The recurrent neural network is embedded into the quasi-linear model,which can be viewed as a quasi-ARX model macroscopically.It is further decomposed into a linear part and a non-linear part.The linear part is identified by a least squares algorithm,and the nonlinear part is identified by a recurrent neural network.Finally,two experiments are used to prove the effectiveness of the identification.Compared with the dynamic extreme learning machine identification,the quasi-linear recurrent neural network has better identification effect.2.Based on the research of dynamic neural network identification,the research of predictive control is studied.However,the dynamic extreme learning machine is prone to overfitting and has a strong randomness and is not suitable for predictive control.Based on these problems,this paper uses quasi-linear recurrent neural network to control the system.In the proposed quasi-linear recurrent neural network predictive control,the recurrent neural network is not directly used as predictive models and controllers,but are indirectly nested in quasi-linear models.What's more,in the quasi-linear recurrent neural network predictive control,the solution of the control law only need one-step derivation,which can greatly simplify the solution process of control law.In addition,the control law of the control law in the predictive control based on recurrent neural network is cumbersome,and the gradient disappears easily.Finally,theoretical analysis and simulations are given to prove the simplicity and effectiveness of the quasi-linear recurrent neural network.