| dc.description.abstract | Thermocatalytic hydrogenation of CO2 to methanol is a promising technology in the fight
against climate change. This process helps to reduce CO2 emissions by transforming it into
chemical compounds such as methanol, which is considered an efficient fuel and serves as a
precursor in chemical synthesis. To improve the process, new materials and operating
conditions are required that can be made available within reasonable time frames, so the use
of artificial intelligence for this purpose can provide notable advantages. In this context, this
study allowed for a descriptive, predictive and causal analysis of the thermocatalytic
synthesis of methanol. For this purpose, a database made up of 3,011 experimental points
obtained through the review of 160 scientific articles. The descriptive analysis revealed that
the process is thermodynamically restricted, so it depends on both the reaction conditions
and the influence of the catalyst. For predictive analysis of methanol space time yield (STY)
from experimental descriptors, five artificial intelligence algorithms were evaluated (Random
Forest, XGBoost, Neural Networks, k-Nearest Neighbors, and Supporting Vector Machines).
The XGBoost and Random Forest algorithms obtained the highest cross-validation
coefficients of 0.881 ± 0.013 and 0.862 ± 0.014 respectively. Once the SHAP algorithm was
applied, it was identified that the most important descriptors in XGBoost and Random Forest
were gas hourly space velocity (GHSV), pressure (P) and reaction temperature (T). | |
