Multivariable Impulse Response Estimation via Correlation Analysis and Its Application to Automated System Identification

Abstract

This paper presents a correlation analysis approach for estimating impulse response coefficients of multivariable systems where inputs and/or measured disturbances are allowed to be correlated with each other. The resulting estimator forms a simple linear equation set for which a simple closed form solution can be obtained, and due to the nonparametric nature of the estimator, the problem of selecting proper model structures encountered in parametric methods is avoided. These characteristics make the estimator very appropriate as a preliminary estimation step to more detailed parametric methods. Two other applications of the correlation analysis estimator include non-zero transfer function determination in multivariable systems and online identification experiment monitoring. The effectiveness of the proposed methods is illustrated on industrial data.