Machine learning-driven chemical processes

 

  This week, most of our work has been completed. For the data preprocessing, we first selected only the parts of the catalytic process that only use plasma, then removed columns that had remained constant, such as reac Ar, N2, etc., and also removed some columns that had too many missing values (more than 50%). Finally, we fill in the missing values of the remaining data, for example, the temperature parameter, we can find that among the existing data, 100 is the most common number, other numbers are not regular and very scattered, so in this column we choose to fill in the missing temperature with the plural 100. Siyuan finally completed the model of neural network. The next thing we need to do is to make a poster.

  It's a new week and we've got reaction-related data, a total of 1180 entries, each with about 20 parameters. Through investigation and questioning we learned the direction of optimization, first we need to select a group of plasma combinations that make the conversion of CO2 and CH4 high, and then analyze the effect of other parameters on the reaction in the selected plasma combinations. So as a first step we want to find the plasma combination that we want to study by means of machine learning. At the same time, Zi'ao needs to find relevant literature to support our next choice of which parameters to explore for the effect on the reaction. For the machine learning models, we roughly determined that Qingyue is responsible for the random forest model, Siyuan builds a BP-neural network, and Haoyu prepares an SVM model.

This week we had an initial look at plasma-catalyzed chemistry and divided up the work among each of us, with Zi'ao doing a survey of the chemistry and plasma-related literature and the rest of us considering which machine learning models to use to optimize the reactions.