(340k) Subspace Model Identification and Model Predictive Control of an Industrial Ethylene Splitter

Jalanko, M. - Presenter, McMaster University
Mhaskar, P., McMaster University
Mahalec, V., McMaster University
This works aims to improve the control performance of an industrial ethylene splitter. The current control strategy uses PI controllers and has no direct end composition controllers causing ethylene products composition to fluctuate due to feed rate and feed composition disturbances. There are several literature studies that consider the implementation of model predictive control (MPC) on distillation columns to improve control performance. These works consider different MPC configuration which can be summarized as direct MPC implementation or cascade MPC implementation [1] [2]. The objective of this work is to investigate the impact of implementing MPC at the supervisory level (cascade configuration) on the control performance of a real-life ethylene splitter. The cascade configuration of MPC and PI controllers is designed so that the PI controllers’ setpoints are determined via the MPC. A linear subspace identification method is adopted to identify a linear state-space model for use within the MPC. This data-driven method is based on gathering the input-output trajectory of the process to construct a linear model [3]. An online model adaptation scheme for the state-space model is developed to improve the model prediction capability under new operation patterns which can improve control performance. To demonstrate the benefits of our work on the real plant, a simulation model that replicate the real-life design is built in ASPEN Dynamics is used as a test bed. Results showing improved performance under the MPC will be presented.


[1] Huang, H., and Riggs, J. B. (2002). Including levels in MPC to improve distillation control. Industrial & engineering chemistry research, 41(16), 4048-4053.

[2] Bezzo, F., Micheletti, F., Muradore, R., and Barolo, M. (2005). Using MPC to control middle-vessel continuous distillation columns. Journal of Process Control, 15(8), 925-930.

[3] Moonen, M., De Moor, B., Vandenberghe, L., and Vandewalle, J. (1989). On-and off-line identification of linear state-space models. International Journal of Control, 49(1), 219-232.