Data cleaning for dynamic modeling and control

"Outliers" or "anomalous data points" occur frequently in practice and can have devastating effects on process data analysis, empirical modeling, or controller implementation. This paper briefly examines the nature of these anomalous data points, their influence, and three possible approaches to dealing with them. One of the key points of this paper is that effective procedures for dealing with outliers must generally be nonlinear. Three different dynamic analysis problems are examined, one based on real process data and the other two based on simulation data for which the exact results are known.