Gross error management in data reconciliation

Abstract This paper reviews the problems associated to the presence of gross errors in data reconciliation problems as well as the main approaches to avoid their undesirable effects: detection and removal of the faulty measurements and minimization of their effect on the reconciled results by using robust estimators that modify the cost function. In the first case, Principal Component Analysis is used to detect the gross errors while, in the second, the Fair function is used. A combined approach is also presented that improves the PCA results. The methods are evaluated in a realistic large-scale problem with plant data corresponding to the hydrogen network of a petrol refinery.