New Identification Algorithm Based On Partial Least Absolute Deviation

With the development of modern control theory,the structure of actual industrial systems is becoming more complex,such as nonlinear systems,closed-loop systems,and multi-variable systems.The study of identification algorithm for complex model is catching more and more people's attention.Considering the correlation of system variables and the target function,the identification method of Hammerstein model,closed-loop model and multivariable model is studied respectively in this paper.For Hammerstein model with colored noise,the extended stochastic gradient algorithm based on approximate least absolute criteria is given,the parameters of noise model can be identified by the proposed method,and the method is suitable for online identification.During the algorithm derivation,the inertia term that contains information of past is introduced for improving recognition algorithm accuracy,and the convergence of the algorithm is analyzed.For the closed-loop system which model order of forward channel less than that of feedback channel without noise,the new identification method of closed-loop system based on partial approximate least absolute deviation and approximate least absolute deviation is proposed respectively.By the proposed methods,the correlation among each measurement data which influence the identification result could be eliminated,and analyzed the identification performance of the two methods.For multivariable system,the coupled identification method based on approximate least absolute deviation is proposed.The algorithm has good calculation speed by avoiding the inversion of high-dimensional matrix and with good robustness,but can not effectively identify the parameters of multivariable systems with correlation between the inputs.Therefore,a new identification method of multivariable system based on partial approximate least absolute deviation is proposed.In this paper,the identification performance of the proposed methods is verified by simulation after the derivation of method in each part of this paper.In the simulation,the comparison between proposed methods and the algorithms based on least squares criterion is given with white noise and impulse noise respectively.Furtherly,the impulse noise with different intensity is used to verify the robustness of the proposed methods,which reflects the effectiveness of the algorithms fully.