Modeling And Parameter Estimation Of A Class Of Delay Systems

Modeling is a very important part of control system design and research,and it is the basis for control system algorithm design and system performance analysis.Generally speaking,methods for establishing mathematical models based on actual operating variables and measurable data in complex industrial sites are divided into two categories: system identification based on model parameter estimation and model-free neural networks.System identification is a gray box modeling method.It is necessary to give the structure and order of the system model on the basis of mechanism analysis,and then identify the corresponding model parameters through process data.The BP neural network modeling method is not the case.This method does not require too much prior knowledge of the system.It only needs to know the input and output data of the system to model the complex system.The change of working conditions is a practical problem often encountered in the operation of industrial systems.The model prediction of a control system under the changing conditions of working conditions is a subject of practical engineering significance in the research of industrial control system modeling.The state space model is the basis of modern control theory.It can describe the dynamic system with delay in the actual industrial field.It takes the single-step delay state space model,the multi-step delay state space model and the multi-delay state space model as objects,and the research has the smallest recurrence.The prediction output and error of each model system under the change of working conditions by the square identification method,the least square identification method of auxiliary model and the BP neural network modeling method.First,according to the input and output data of each model system under stable conditions,the system model parameters are identified by the recursive least square method and the auxiliary model least square method.At the same time,the BP neural network method is used to construct the system approximation model.Then compare and analyze the system prediction output and error of the identification model and the neural network model;then simulate the changes of the actual industrial site by changing the boundary conditions of the operating variables,and give a comparative simulation of the three methods.The research results show that the recursive least squares method,auxiliary model least squares method and neural network modeling method can all simulate and track the system with known model structure;the predicted output of the three model systems under stable conditions In terms of error,the BP neural network modeling method is better than the auxiliary model least squares identification algorithm and the recursive least squares method;but under the condition of changing working conditions,the BP neural network enters a state of relearning Therefore,its followability to the system model output becomes worse,and the model output error performance is inferior to the auxiliary model least square identification method.