Using Sub-models for Dynamic Data Reconciliation

The optimal approach for dynamic data reconciliation consists in using a complete and exact process model. Unfortunately, such a model is difficult to obtain in industrial practice. Through an example, several observers based on static, stationary and dynamic sub-models are designed and compared to the optimal approach. The comparisons illustrate that, for the given conditions, static observers generally lead to estimates that are less precise than the measurements. Stationary observers are slightly more precise than static observers but they obviously lack the power of temporal redundancy offered by dynamic models. Deterministic dynamic sub-models, that do not include all physical parameters (thus relatively easy to obtain), which stochastic models are added to, are shown to give good performances. Copyright © 2006 IFAC