Data-Driven Discovery of Reaction Pathways for Modeling\nCatalytic Cracking of Hydrocarbons

Modeling\ncomplex chemical reactions is a long-standing challenge\nin chemical reaction engineering. Group type analysis is commonly\nused for reducing the complexity of the reaction model while retaining\nsufficient chemical information for a particular application. This\nstudy evaluates a data-driven approach based on mathematical similarity\nas a new tool for identification of groups for use in group type reaction\nmodels. Data for dodecane cracking in the supercritical state over\nZeolite Socony Mobil-5 (ZSM-5) was used as a test system. Mathematical\nsimilarity analysis of the raw data differentiated as many as five\ndifferent groups. Adding synthetic data did not affect the groups\nidentified by similarity analysis, indicating that the separation\nwas not limited by the number of data points. Scaling or normalizing\nthe data improved the separation of the similarity analysis. To test\nthe data-identified groups, different types of reaction models were\ngenerated systematically, kinetic parameters regressed, and the resulting\npredictions compared with experimental data. The resulting reaction\nmodels consisted either of parallel or sequential reactions. As a\ngeneral statement, the reaction models consisting of parallel reactions\nwere more accurate than those consisting of sequential reactions.\nAdditional tests showed that user defined groups could be added to\nthose identified mathematically to improve the accuracy of predictions\nfor target species without sacrificing overall accuracy. The similarity\napproach was applied to a data set consisting of catalytic dodecane\ncracking in the presence of water added to the reaction mixture (50\nwt %). The proposed similarity analysis identified different groups\nin the presence and absence of water, indicating that the data-driven\napproach can be used to identify qualitative differences in the reaction\npathways. It is therefore demonstrated that data-driven identification\nof group types represents a useful new tool for development of group\ntype models.