Development of a new comprehensive predictive modeling and control framework for multiple-input, multiple-output processes

摘要

The increase in the competitiveness of chemical process industries has necessitated the need for lowered energy and raw material consumption and improved quality control with tighter limits. This stronger control of the process conditions has generated interest in the use of advanced process control, and Model Predictive Control (MPC) is one such approach. The idea behind MPC is the use of a model to predict future behavior and use that knowledge to manipulate the input variables so that a cost function is minimized, resulting in optimal control. The predictive model is the core of the MPC method, and the success of a particular strategy hence depends on the accuracy of the model. The task of model building is also very challenging and time-consuming, so an ideal modeling approach would be one that does not require too much data but maintains its accuracy. The semi-empirical modeling approach has the strength that by using an intelligent model form, it has minimal data requirements. The use of semi-empirical models was first demonstrated by Rollins et al., and they coined the term SET (semi-empirical approach) for their method. The accuracy of SET over conventional empirical models was one of the biggest advantages of that approach. The ease of model identification, robustness in parameters, and a novel algorithm were some of the other strengths. Thus SET showed great potential for use in a multivariate situation, and the results of this work have shown that SET has been able to handle this challenge successfully. This extension led to the creation of a comprehensive modeling framework, which is far superior to the current modeling approach using the semi-empirical models. The various components of this approach are: use of statistical design of experiments, model identification, a novel model structure, and finally an algorithm that seeks to maximize accuracy.