Multivariable Model Identification Method Without External Test Signal

The frequency response usually contains the dynamic characteristics of a system,so study on algorithm in frequency domain increasingly draw researchers' attention,especially in multivariable identification domain.Since the analytic expression of frequency response is not easy to find,the common way is to apply an external test signal to system,then obtain the frequency response from its response curve.However,confined to security limits,it's not always possible to do signal test.Based on input-output data of system,the frequency domain identification algorithm of multivariable system with no external test signal is studied in this paper.The normal operating data of multivariable system from one steady state to another with no external test signal are selected for identification in the paper.In order to do Laplace transformation of the measured signal,we first do a difference operation of the measured signal,then divide the whole sampling time into several time periods and make an approximation of the difference signal on each time period,so the Laplace integral sum in the whole sampling time can be transferred into a numeric calculation expression,and the frequency response of the system can be get.Choosing a second-order model with time delay for model matching,through magnitude criteria and angle condition,liner equations with to-be-identified parameters are derived.Then complete least square method,liner regression method based on least absolute deviation criteria and recursive algorithm based on approximate least absolute deviation are deduced to get the transfer function model of multivariable system.Furthermore,algorithm improvement for non-minimum phase system is discussed.By changing the magnitude criteria and angle condition accordingly,the process model of non-minimum phase system could be obtained too.Simulation experiments prove the correctness and excellent calculation performance of the method proposed.Identification results of data disturbed by noise show the good robustness of the method.More precise identification results of disturbed data can be got after de-noising processing.