A Generalized Machine Learning Framework for Understanding and Optimizing Thermocatalytic Conversion of Syngas into C5+ Hydrocarbons Fuels

Abstract

Thermocatalytic conversion of the renewable syngas into long chain hydrocarbons fuels was an attractive energy production technology, for combating climate change, energy crisis, and wastes disposal. However, this thermochemical process was very complicated, and target product also highly depended on the feedstock information, catalyst properties, and process condition. At present, it was still challenging to fully understand and optimize this process. To address this gap, we developed a machine learning framework to model Fischer-Tropsch synthesis process of syngas towards C5+ hydrocarbons fuels from experimental descriptors. A database of Cobalt-based catalyst with 406 datapoints was compiled from literature and subjected to data mining. Accurate ensemble-tree models (R2 > 0.82) were developed to predict the CO conversion and C5+ hydrocarbons fuels selectivity from 12 descriptors, where the significance of dispersion, pressure, temperature, and metal content was revealed. Casual analysis revealed that C5+ hydrocarbons fuels selectivity was positively correlated with lower temperature (< 481 K) and higher dispersion (> 7.72%). Besides, some interesting findings were also observed, for example, smaller cobalt size, and lower pore size (< 9.27 wt%) and cobalt loading (< 22 wt%) were positively related to C5+hydrocarbons fuels selectivity. The framework was purely data-driven, interpretable, and highlighted the ability of this method to unearth relationships of target variables and descriptors in thermocatalytic conversion of syngas, by isolating effects of individual design parameters in a manner that would be difficult to achieve experimentally. These insights could help design and optimization of subsequent experiments.

Keywords: Fischer-Tropsch synthesis, machine learning, CO conversion, C5+hydrocarbons selectivity