Multiple data sets for model updating in real-time operations optimization

Abstract

This paper presents a method for improving model-updating methods used for the real-time optimization (RTO) of plant operations. Previous work on updating and results analysis and diagnosis is extended to the use of multiple data sets for updating the steady-state plant model and to use prior knowledge to categorize the parameters as fast or slow changing. A key challenge in real-time estimation is identifying the maximum number of parameters that can be estimated reliably using the current data; this number changes due to differences in variability in operating condition and sensor availability. Since reliable parameter estimation yields a parameter covariance matrix with a small condition number and determinant, the number is based on a real-time diagnostic that is based on the covariance matrix. Case studies demonstrate the importance of a real-time updating diagnostic and indicate that when plant variation exists in multiple data sets, increased profit can be obtained by updating the additional parameters to reduce the plant/model mismatch.