Research On Accurate Model Identification Method Of Electromechanical System Based On Radial Basis Function Neural Network

In this thesis,we focus on the precise modeling method for the servo system.By analyzing the complex and imprecise problems of mechanism modeling,we point out that if using the neural networking modeling method,the rapidity,accuracy and simplicity canbe improved significantly.Nowadays many research working on neural network identification and many improvement schemes can be found,however,most of them are only effective for some specific simulation models and not test for the real systems.Therefore,this paper focus on the performance improvement of the neural network based model identification of the actual servo system,the main results can be summarized as follows:First,a nominal model analysis and detailed perturbation analysis are carried out for a class of position servo system which uses the permanent magnet synchronous motor as the actuator.And it is pointed out how the different nonlinear parts and perturbation items influence the neural network identification.Secondly,the basic structure of neural network identification and the basic method of neural network training are compared and analyzed.Thirdly,according to the characteristics of servo system,the two-point differential series-parallel identification structure is proposed and neural network structure for the given servo system is determined.Then combined the Orthogonal Least Squares(OLS)and Gradient Descent method(GD),this training algorithm is designed to optimize the number of neural nodes and the location of the neural network bias function central points.Then inspired by the servo system sweep method,we given the selection scheme of the sample data,test data and design the evaluation target function.At last the effectiveness of the improvement is verified by some simulation experiments.Finally,combined with the designed neural network based model identification scheme of the servo system,the open-loop training samples and test data are collected in the actual turntable servo system,then one neural network model with these data are trained.And through contrast with the traditional sweep identification method,it can be seen that the feasibility of neural network for modeling actual servo system.