Data-driven robust model predictive control technology for propylene distillation process

Abstract

The distillation column is always affected by external disturbances during its operation. Using data-driven robust model predictive controller (DDRMPC), which based on the data-driven robust optimization (DDRO) method, can better handle the process uncertainty than the traditional robust model predictive control (TRMPC) because of the introduction of the machine learning method. A DDRMPC of propylene distillation column is proposed to hedge against the uncertainty of propylene content at the top of the column. Firstly, a linear state space model of the process is established based on the compartmental method and the dynamic mechanism model, and then the uncertainty set of principal component analysis and robust kernel density estimation is constructed by using the historical data. Certainty equivalent MPC (CEMPC), TRMPC and DDRMPC algorithms are constructed respectively. Finally, the performance of DDRMPC is analyzed through the case study of composition control.