Model Deficiency Diagnosis and Improvement via Model Residual Assessment in Model Predictive Control

• Published in: Industrial & engineering chemistry research Vol. 56; no. 42; pp. 12151 - 12162
• Main Authors: Li, Lijuan, Song, Jianquan, Zhang, Xiaoxiao, Ye, Jing, Yang, Shipin
• Format: Journal Article
• Language: English
• Published: American Chemical Society 25.10.2017
• ISSN: 0888-5885; 1520-5045
• DOI: 10.1021/acs.iecr.7b02598

To reduce the effort and cost of model maintenance in model predictive control (MPC) systems, this paper explored a model deficiency diagnosis and improvement method by the assessment of model residual and optimization of disturbance model. A model quality index (MQI) method was first presented to evaluate the model performance with the routine input and output process data. Based on MQI, a leave-one-out method was proposed to further assess the performances of submodels in multi-input–multi-output (MIMO) MPC processes. The root deficient submodel could be diagnosed based on a comparison of the overall MQI with the submodel index by moving it to the disturbance channel. Further, a model performance improvement method through the upgraded optimal disturbance model was proposed to enhance the model performance without altering the control model. The experiment results on the Wood–Berry distillation column process and an industrial process indicated the validity of the proposed model diagnosis and improvement method.