A kind of modeling method based on the Bayesian dynamic linear model for model predictive control of a nitrobenzene prefractionator

Abstract

The plant testing method, which has the disadvantages of time consuming and production disruption, is a usual system modeling method in process industries. This paper presents the Bayesian dynamic linear model (DLM) using the subjective experience information from operator and fewer step tests to calculate and forecast the model for model predictive control (MPC), which is used to control the bottom stage temperature of a nitrobenzene prefractionator. The effectiveness of the model based on the Bayesian DLM is illustrated by a comparison of predictive effects with the model identified by step tests, and a MPC controller using a model based on the Bayesian DLM with eight step tests and initial experience information is designed and put into operation. The practical application shows that the designed MPC controller could effectively improve the stability of the bottom temperature and the system modeling method based on the Bayesian DLM could reduce the number of step tests and has great economic and practical significance.