OUTLIER DETECTION IN-PROCESS PLANT-DATA

An integrated approach for outlier detection and data reconciliation i s discussed. It is shown that outliers can be identified by directly e xamining the measurement distribution. In our approach, a non-linear l imiting transformation which operates on the data set is utilised to e liminate or reduce the influence of outliers on the performance of the conventional data reconciliation. Monte Carlo study shows that the pr oposed approach provides better results than the conventional approach in the presence of outliers. When no outlier exists, it provides as g ood a performance as the conventional approach. (C) 1998 Elsevier Scie nce Ltd. All rights reserved.