Robust Identification Using Semiquantitative Methods

Abstract While model-based process monitoring and system identification share the common goal of describing the behavior of a physical system based on a mathematical model, the assumptions made by system identification are typically too strong to apply it directly to monitoring problems. This article describes SQUID, a new system identification method that uses refutation rather than search to identify a model of a physical system. By ruling out implausible models rather than searching for the best model that fits the data, SQUID is more robust in the face of uninformative data and structural model uncertainty than are traditional identification methods.