Detection of Significant Model-Plant Mismatch from Routine Operation Data of Model Predictive Control System

The maintenance of model predictive control (MPC) systems is one of the major problems identified by industrial process control engineers. Since performance deterioration is usually caused by changes in process characteristics, effective re-modeling is the key to success. Obviously, not all sub-models have to be reconstructed; thus, it is crucial to identify sub-models that have significant model-plant mismatch. In the present work, a novel method is proposed for significant model-plant mismatch detection from routine closed-loop operation data on the basis of the statistical test concept. The effectiveness of the proposed method is demonstrated through case studies. The results clearly show not only that the proposed method can detect sub-models that have significant mismatch but it is superior to the other methods based on multivariate analysis.