Data-driven Identification of Stoichiometric and Kinetic Models for Complex Reaction Mixtures
Fromer, Jenna C.
2021
-
Target Factor Analysis (TFA) quantifies whether a hypothesized candidate stoichiometry is in agreement with time-resolved composition data from a reacting mixture. This has been recently aided by the Dynamic Response Surface Modeling (DRSM) methodology (Klebanov, 2016), where continuous concentration profiles are estimated from discrete compositional measurements. In the present thesis, we remove ... read morethe requirement to postulate candidate stoichiometries. We define an algorithm for the identification of stoichiometric and kinetic models that accurately model the reaction mixture data. Initially, all possible reaction stoichiometries that satisfy the mass balance constraint and pass the TFA test are identified. The resulting stoichiometric candidates are combined to form full rank candidate reaction networks of multiple reactions each. Several of these networks are filtered out through additional constraints. Kinetic models of different forms are estimated separately for each reaction of the remaining reaction networks. The accuracies of competing reaction networks are compared among themselves using F-tests of the corresponding regression sums of squares. We subsequently test if all non-random data are represented by the most accurate network. Three case studies have been analyzed involving three, six, and eleven species, respectively. For the first two systems, involving two and three reactions, respectively, the true stoichiometric and kinetic models were identified. For the complicated case of eleven species and eight reactions, the obtained models from the first pass of the proposed algorithm fail to represent all non-random data. This necessitates the need of a second modeling cycle.
Advisor: Professor Christos Georgakisread less - ID:
- 1544c366z
- To Cite:
- TARC Citation Guide EndNote
- Usage:
- Detailed Rights
