Limited Time Offer

Claim a 25% discount on all eLearning courses (including credentials) with code ELEARN25.

Offer is valid from March 10-31. Public courses excluded from promo. 

(470c) Adaptive System Identification of an Industrial Ethylene Splitter: A Comparison of Subspace Identification and Artificial Neural Networks

Jalanko, M. - Presenter, McMaster University
Mhaskar, P., McMaster University
Mahalec, V., McMaster University
The work considers the problem of data driven modeling of the operation of an ethylene splitter (and in particular, the ethylene splitter at the Joffre site in Alberta) used to separate ethylene from ethane. The process exhibits highly nonlinear behavior and presently operates with no direct end composition controllers making the process sub-optimal in terms of the products purities and energy consumption. The two types of data-driven models that can be obtained from process input-output data are input-output model and state-space model [1]. State-space model such as subspace identification models have been developed for distillation column for control purposes [2] [3]. Also, different neural networks models have been used to develop a model for distillation column [4] [5]. The objective of the present work is to investigate the use of different data driven techniques such as linear subspace identification, nonlinear feed forward neural network, and nonlinear recurrent neural network for the purpose of developing a dynamic data driven model. To this end, first an ethylene splitter simulation model is built that replicates industrial operation. The ability of the simulation model to capture the key traits of the process dynamics are first established by comparing with data from plant operation. The simulation model is subsequently utilized to generate data and utilize to test the ability of the modeling tools for interpolation and extrapolation. An online model adaptation scheme is developed for some of these methods to improve their prediction capabilities under new operation patterns. The methods are finally utilized to model (and are validated against) industrial data.


[1] Phan, M. Q., and Longman, R. W. (1970). Relationship between state-space and input-output models via observer Markov parameters. WIT Transactions on The Built Environment, 22.

[2] Castaño, J. E., Patiño, J. A., and Espinosa, J. J. (2011). Model identification for control of a distillation column. In IX Latin American Robotics Symposium and IEEE Colombian Conference on Automatic Control, 2011 IEEE (pp. 1-5). IEEE.

[3] Meenakshi, S., Almusthaliba, A., and Vijayageetha, V. (2013). MIMO Identification and Controller design for Distillation Column. International journal of innovative research in electrical, electronics, instrumentation and control engineering, 1(2), 44-48.

[4] Kanthasamy, R., Anwaruddin, H., and Sinnadurai, S. K. (2014). A new approach to the identification of distillation column based on hammerstein model. Modelling and Simulation in Engineering, 2014.

[5] MacMurray, J., and Himmelblau, D. (1993). Identification of a packed distillation column for control via artificial neural networks. In 1993 American Control Conference (pp. 1455-1459). IEEE.