Variational Bayesian Identification Method Based On Oversampling Closed-loop Structure

According to the traditional closed-loop identification theory,it is generally necessary to add additional test signals or increase the order of the controller beyond that of the controlled object to obtain data that satisfies the identifiability.However,in actual industrial processes,the order of the controller is not always larger than that of the controlled object,and the addition of test signals will not only cause huge identification costs,but even affect the normal operation of industrial processes.In view of the shortcomings of adding additional test signals,this paper uses the advantages of oversampling closed-loop structure without adding any additional test signals,and adopts the variational Bayesian method with strong adaptability and high accuracy as the identification method,and variational Bayesian methods based on oversampling closed-loop structure under the condition of different noise environments and different identification models were proposed in the paper.Firstly,on the basis of the over-sampling closed-loop structure,the variational Bayesian method based on oversampling closed-loop structure is proposed for the identification of linear models under white noise and complex noise.The variational Bayesian method based on oversampling closed-loop structure not only overcomes the shortcomings of traditional closed-loop structure identification,but also has higher identification accuracy.At the same time,by analyzing the asymptotic variance expression of the estimated model,the oversampling structure identification can use the high frequency information of the extra noise to improve the identification accuracy when the output noise is polluted by impulse noise or spike noise.Then,on the basis of the variational Bayesian method and oversampling closed-loop structure,a variational Bayesian method based on oversampling structure is proposed for multivariable nonlinear model identification.The proposed multivariable variational Bayesian method also has the advantages of improving identifiability and high recognition accuracy.At the same time,in order to make the variational Bayesian method applicable in the case of colored noise,the original variational Bayesian method is improved,and the multivariable amplitude limiting variational Bayesian method based on oversampling structure is proposed which improved the convergence performance of the proposed method when the colored noise exist in the measured data.Finally,for the identification of a class of process objects polluted by outliers,using the student-distributed noise to describe the process contaminated by outliers,a robust variational Bayesian identification method suitable for the student-distributed noise is proposed,and the proposed method also combined with the oversampling closed-loop structure.The robust variational Bayesian method based on oversampling closed-loop structure not only has smaller identification error when the influence of outliers is small,but also maintains high identification accuracy when the outliers have a large influence.