(187f) Reaction Identification and Parameter Estimation from Chemical Process Data

Wilson, Z., Carnegie Mellon University
Sahinidis, N., Carnegie Mellon University
Mixed-integer non-linear programming (MINLP) has been utilized to facilitate model reduction in chemical reaction networks in the literature [1, 2]. Inspired by recent success in the application of mixed-integer programming to the domain of surrogate model development [3], we propose a MINLP methodology to identify the reactions, mechanisms, and estimate the associated kinetic rate parameters of a reaction network from chemical process data. Application of this approach to dynamic process data requires the use of algebraic surrogate models for the concentration or conversion profiles [4]. This approach utilizes a superset of chemical reactions, which can contain different mechanistic pathways for the same reaction. A model-building procedure, similar to one described previously for subset selection in multiple linear regression [3], is then used to minimize an objective that balances the complexity of the model with its fit in order to identify a probable subset of reactions and estimate the associated kinetic rate parameters. The utility of this methodology is demonstrated on a number of steady-state and continuous computational problems, as well as an experimental data set that affords the consideration of different oxidation mechanisms in a chemical looping combustion reactor.

References cited

[1] Edwards, Keith, T. F. Edgar, and V. I. Manousiouthakis. "Reaction mechanism simplification using mixed-integer nonlinear programming." Computers & Chemical Engineering 24.1 (2000): 67-79.

[2] Bhattacharjee, Binita, et al. "Optimally-reduced kinetic models: reaction elimination in large-scale kinetic mechanisms." Combustion and Flame 135.3 (2003): 191-208.

[3] Cozad, A., N. V. Sahinidis, and D. C. Miller, Automatic learning of algebraic models for optimization, AIChE Journal, 60, 2211-2227, 2014.

[4] Wilson, Zachary T., and Nikolaos V. Sahinidis. "The ALAMO approach to machine learning." Computers & Chemical Engineering, accepted (2017).