Wavelet-Based Multiscale Identification For A Stochastic CARMA Model

Recently, Control-related process identification method has been widely studied. Process identification associating with the control performance index raises new requirements for traditional identification methods, Using the wavelet multiresolution analysis identification is a control-related process identification method. Based on discrete wavelet transformation, the input and output data of the process is decomposed into different frequency bands, accurate modeling in the operation band of the process can be achieved by using the data related or the weighted data of the different frequency bands.Stochastic CARMA model is fundamental for the control process modeling. In this paper, multiscale identification for stochastic CARMA model is discussed, which is organized as follows:1. Based on Multistep Least-Squares(MLS) system identification, a method of multiscale parameter identification for Instrumental model and process model is presented;2. Deriving the equivalent time-domain models (M_j|(—o)) form the estimated multiscale models M_j and identifying the parameters of the noise model in the time-domain;3. Selecting the optimum model M from (M_j|(—o)), which meets the demands of the control system;4. The valid conclusions of the presented method in this article is made by simulation examples.The work in this paper has proved that Wavelet—based multiresolution analysis is utilized to extract information at different frequency scales from the data of the process estimated, and to enhance the signal to noise ratio and thus to improve identification accuracy. It facilitates relating the requirements of the model for the control system design to the identification procedure.