Research On The Multivariable System Identification Based On Closed-loop Condition

It's meaningful to study multivariable system identification theory in closed-loop condition and take it into practice.In the recent years,subspace identification algorithm has already been a significant subfield in the system identification field.The method uses the input and output data to estimate parameters of system state space,which is an effective means for multivariable system modeling and analysis.This paper introduces the basic principles and classical algorithms of subspace identification method.Considering the fact that the classical subspace identification method is not appropriate in the closed-loop situation,based on the errors-in-variables model,the paper proposes a subspace identification algorithm combined with principal component analysis.Instrumental variables method is recommended to eliminate the influence of noise,which avoids the projection calculation.An improvement is proposed during the principal component analysis process,which simplifies the solution procedure.In the last,taking unit plant coordinated control system's dynamic characteristics into account,simplifies it as an unknown model with two inputs and two outputs,and uses the real-time data to estimate matrixes of state space system.And then,use the data from disturbance test to check the model we have got,and compare the results with the MEOSP method.The results shows that the subspace identification method based on principal component analysis is a relative accurate algorithm for estimating parameters of MIMO system,which will be reference for practical engineering.