(251c) Detection of Significant Model-Plant Mismatch Using Stepwise Method for Model-Based Control | AIChE

(251c) Detection of Significant Model-Plant Mismatch Using Stepwise Method for Model-Based Control


Shigi, Y. - Presenter, Kyoto University
Kano, M. - Presenter, Kyoto University

The results of a questionnaire to member companies of the JSPS (Japan Society for the Promotion of Science) 143rd committee on their process control applications clarifies that model predictive control (MPC) has been widely and successfully implemented in the chemical and petroleum refining industries but problems still remain to be solved [1]. To achieve desirable performance, it is necessary to build an accurate model and to tune control parameters appropriately. However, both of them are difficult in practice due to process nonlinearity and changes in process characteristics. To keep sufficient control performance and to prevent or at least cope with performance deterioration, the maintenance of MPC is crucial. Control engineers need to know the reason of performance deterioration and the effective countermeasure.

In MPC, manipulated variables are determined on the basis of prediction results of process responses by using a plant model. Therefore, the control performance of MPC depends on the accuracy of the model. When the performance degradation is caused by model-plant mismatch, it should be solved by re-identification of the model. However, for processes with a large number of inputs and outputs, re-identification is very laborious. Badwe et al. [2] proposed a methodology for the detection of model-plant mismatch, elements of a transfer function matrix that have significant residuals, from closed-loop operating data using partial correlation analysis between the model residuals and the manipulated variables, and they demonstrated the efficacy of their method through the two simulation case studies. However, the following questions remain unsolved. Does the proposed detection method always function well? Is there any limitation on its applicability? What should we pay attention to when detecting plant-model mismatch? These questions need to be solved from a practical viewpoint.

In this work, a theoretical analysis of the model-plant mismatch detection method based on partial correlation analysis was conducted. The results demonstrate that the partial correlation coefficients between model residuals and manipulated variables largely depend on the variances of set-points and disturbances. Therefore, it can be said that the model-plant mismatch is serious when the partial correlation coefficient is significant; but the converse is not always true. In other words, the model-plant mismatch might be serious even when the partial correlation coefficient is small. Thus, the model-plant mismatch detection method based on partial correlation analysis should be used carefully. Furthermore, in the present research, the theoretical results are supported by simulation results, and extension of the existent method is proposed to improve the reliability of the model-plant mismatch detection.

[1] M. Kano and M. Ogawa: The State of the Art in Advanced Chemical Process Control in Japan, IFAC ADCHEM, Istanbul, Turkey, July 12-15 (2009)

[2] A. S. Badwe, S. L. Shah, S. C. Patwardhan, and R. S. Patwardhan: Model-Plant Mismatch Detection in MPC Applications Using Partial Correlation Analysis, IFAC World Congress, Seoul, Korea, July 6-11 (2008)


This paper has an Extended Abstract file available; you must purchase the conference proceedings to access it.


Do you already own this?



AIChE Pro Members $150.00
AIChE Graduate Student Members Free
AIChE Undergraduate Student Members Free
AIChE Explorer Members $225.00
Non-Members $225.00