ML based catalyst modelling for photo-catalytic reduction of CO2 to methanol

Abstract

Industrial revolution & globalization lead to the rampant increase in the anthropogenic CO2 emissions which has become the global concern. Photo-reduction of CO2 to other value-added products such as methanol can be observed as a technology that has drawn immense attention of numerous researchers these days. This work majorly focuses on developing the best machine learning (ML) model that can predict the formation rate of methanol from CO2. Present study considers the development of several ML models such as Linear Models; Tree based models and Neural Network Models wherein the extensive data mining (505 data points, from last decade) lays a great foundation. In this process of mining the data, 21 different experimental parameters have demonstrated huge influence on the methanol formation rate from CO2. On comparing their performances, it was observed that Neural Network models (R2 ∼ 0.70) were the most promising models that have exhibited the best performance amongst other ML models. Apart from obtaining the best model this work also throws light on the quantifying influence of experimental parameters (known as input contribution) which aids in arriving at the crucial parameters that can be tuned to accelerate the performance of the model. It was also observed that the catalyst composition, reaction temperature, source of light and its wavelength have created a huge impact on the formation of methanol.