A Unified Approach to Identification of Linear SISO Models Subject to Missing Output Data and Missing Input Data

摘要

When output data is missing in a system identification scenario, it is not the Euclidean norm of the prediction error vector per se that should be minimized. Doing so will almost always yield biased parameter estimates. Two algorithms for estimation of the parameters, which can handle both missing output data and missing input data, are presented. The criterion minimized in the algorithms is the Euclidean norm of the prediction error vector scaled by a particular function of the covariance matrix of the observed output data. The algorithms yield a maximum likelihood estimate of the parameters under certain conditions.