结合深度神经网络的大气压脉冲放电转化CO_(2)研究

Study on Atmospheric CO_(2)Discharge Driven by Pulsed Voltages Through Introducing Deep Neural Network

为了提高等离子数值模拟在放电等离子体转化利用CO_(2)研究中的计算效率,提出了采用具有多个隐藏层的深度神经网络(DNN)来研究大气压脉冲放电转化CO_(2)的放电特性与等离子体化学性质,经过训练的DNN能够极大地提高计算效率。DNN预测结果表明:当外加电压幅值不变时,增加脉冲上升率可以提高放电电流密度和击穿电压,同时增强鞘层区域的电场;脉冲坪区宽度的增加会提高介质板表面电荷密度,增强反向感应电场的强度,进而提高脉冲下降阶段的放电电流密度。此外,提高脉冲上升率和坪区宽度都会提高CO、O_(2)等产物的密度,导致CO_(2)转化率的增加。基于有限的训练集,DNN能快捷、准确地给出海量的数据以揭示CO_(2)放电的演化特性与转化规律,这为研究放电等离子体技术转化CO_(2)提供了新的计算工具。

To improve the computational efficiency of plasma numerical simulation in the study of discharge plasma for conversion and utilization of CO_(2),a deep neural network(DNN)with multiple hidden layers was proposed to investigate the discharge characteristics and plasma chemical properties of atmospheric pressure pulse discharge for conversion of CO_(2).The well-trained DNN can greatly improve computational efficiency.The DNN prediction results show that when the applied voltage amplitude is fixed,increasing the pulse rise rate can improve the discharge current density and breakdown voltage,and enhance the electric field in the sheath region.Moreover,an increase in the pulse plateau width can improve the surface charge density of dielectric plate and strengthen the intensity of reverse induced electric field,so as to enhance the discharge current density during the pulse fall phase.In addition,increasing the pulse rise rate and the plateau width can improve the density of product particles such as CO and O_(2),thus leading to an increase in CO_(2)conversion.Based on a limited training dataset,the well-trained DNN can rapidly and accurately yield massive data to obtain insight into the evolutionary characteristics and conversion laws regarding CO_(2)discharge,which provides a new computational tool to study the CO_(2)conversion using discharge plasma technology.