A Parameter Recognition Method Of The Nonlinear System Based On Nuclear Learning

Nonlinear system modeling and parameter identification has been a hot topic because of its numerous applications.In recent years,kernel learning method has become increasingly popular,and parameter identification of nonlinear system based on kernel learning method has also been concerned by a mount of scholars.The main idea of kernel learning method is using kernel technique to map input data into high dimensional space,and nonlinear problem can be transformed into linear problem.Finally,linear processing of data can be carried out.Linear calculation by mapping can reduce the calculation amount and improve the identification efficiency.This thesis uses K-OPLS,MK-LSSVM and FVS-KELM methods to identify and compare the parameters of two-dimensional,three-dimensional and four-dimensional nonlinear systems.The main research contents include the following aspects:(1)Research the kernel latent variable orthogonal projection(K-OPLS)method,first train and predict the model,and apply to the reconstruction experiment of two-dimensional nonlinear system,that is,Duffing chaotic system.The identification model image is compared with the original system attractor,Poincare mapping and bifurcation diagram images,and the parameter changes of the identification model image corresponding to the original system are obtained.The invariance index of the model is measured to show the dynamics of the K-OPLS identification method refactoring performance.The results show that the K-OPLS model has a good ability to judge the dynamic invariance index,and the identification effect is good.(2)Research the multi-core least square support vector machine(MK-LSSVM)method,the MK-LSSVM method is applied to the three-dimensional chaotic system Chen chaotic system based on the comparison K-OPLS model and the parameter identification ability of the model.By comparing the identification model image with the original system Poincare mapping and bifurcation image,the parameter change of the identification model corresponding to the original system is obtained.The correlation coefficient identification rate of the identification model in Poincare mapping and bifurcation diagram is obtained by data calculation.(3)Research kernel limit learning machine based on feature vector selection(FVS-KELM)method.The FVS-KELM method is applied to the joint angle identification of four-dimensional nonlinear system,that is,then the FVS-KELM method is applied to the joint angle identification of four-dimensional nonlinear system,that is,manipulator nonlinear system.The results show that the FVS-KELM method has a good effect in the modeling and the learning speed of the model is fast and the identification accuracy is high.To sum up,by comparing the modeling and parameter identification effects of kernel learning based K-OPLS,MK-LSSVM and FVS-KELM methods in two,three and four dimensional nonlinear systems,we can see that the FVS-KELM identification effect is the best,and the MK-LSSVM and K-OPLS methods are the second.