Dielectric Barrier Discharge Plasma-Enabled Energy Conversion Under Multiple Operating Parameters: Machine Learning Optimization

Abstract

Dielectric barrier discharge is an important method in plasma-enabled energy conversion. By coupling different power sources, plasma parameters can be easily controlled by a variety of operating parameters to optimize plasma-enabled non-oxidative methane conversion and plasma-catalytic ammonia synthesis. Due to the complexity of the reactions in the plasma, the application of the trial-and-error experiment method to multi-parameter problems will consume a lot of resources and time. When the cause of the change in response can be known, multi-parameter regression and sure independence screening and sparsifying operator can reasonably predict the changing relationship between the influencing factors and the experimental results, and at the same time give the expression, which is applied to the prediction of plasma-enabled non-oxidative methane conversion under different rising times, pulse widths, frequencies, and voltages. However, catalysts are usually added in plasma energy conversion. The characteristics of catalysts are determined by multiple macro- and micro-characteristics. If fitting analysis is carried out for each feature, the problem of data explosion will be brought about, and this is not feasible in the experiment. Therefore, the artificial neural network is used to explain the influence of the N₂ ratio and gas temperature of different catalysts due to the lack of clear characteristic quantity to characterize the catalytic action in plasma-catalytic ammonia synthesis. Different machine learning methods applied to different problems will accelerate the parameter optimization in plasma-enabled energy conversion.