A new robust direct method for measurement error covariance estimation

Abstract

Estimation of the measurement error covariance matrix is an essential requirement in data reconciliation methods. It is common practice to assume that the measurement errors are normal and have a known covariance matrix. A new robust direct algorithm for measurement error covariance estimation is proposed in this paper. Hampel's three-part redescending M-estimators are used to nullifies the effect of large outliers. A direct scheme treating the measured process variables is adopted to make it be used in the cases of nonlinear constraints. Implementation results show that credible results can be achieved either with or without the presence of external causes.