Determining a Collision Cross-Section Set from Electron Swarm Parameters Using Machine Learning Method

A complete collision cross section set of eco-friendly gases is very important for the study of the micro-discharge mechanism of these gases. According to the self-consistent physical connection between the electron swarm parameters and the collision cross sections, the prediction of the collision cross sections are made through the analysis of electron swarm parameters by pulsed Townsend experiments. To automate this process, we use the machine learning method to establish the mapping relationship from electron swarm parameters to collision cross sections. Firstly, we present a suitable neural network with electron swarm parameters as input and collision cross sections as output. Then, We train the neural network using collision cross sections from the LXCat project, paired with electron swarm parameters calculated by Boltzmann equation. Finally, We successfully apply this machine learning approach to obtain a set of collision cross sections of C₄F₇N gas, that refines the set published in previous study (XJTUAETLab database in http://www.lxcat.net [1]).