Abstract: A datadriven approach for model predictive control performance monitoring was proposed. An overall performance index based on Mahalanobis distance was introduced with its benchmark deduced to achieve higher monitoring performance. To identify the root cause of performance degradation, Mahalanobis distance based performance diagnosis method was proposed. Performance signatures are extracted from both principal component and residual subspace, and a classifier is constructed to identify four common performance degradation patterns. The effectiveness of the proposed method was demonstrated in a case study of the WoodBerry distillation system.

Key words: datadriven, model predictive control, Mahalanobis distance, performance monitoring, performance diagnosis