摘要
研究有害化学气体扩散规律并构建其扩散模型,有助于在突发化学事故条件下对有害化学气体的扩散范围进行快速预测,为采取相应救援行动提供支持。本文基于物理信息约束的神经网络(Physical-Informed Neural Network,PINN)模型和流体力学的Navier-Stokes方程,将卷积神经网络(Convolutional Neural Network,CNN)与长短期记忆(Long Short-Term Memory,LSTM)网络结构相结合,重点对PINN在有害气体预测过程中的网络结构构建、网络算法流程以及需要解决的关键问题进行了初步研究,设计搭建了基于PINN的模型基本结构;在此基础上,提出了N-S方程适用性、PINN模型驱动方式等需要进一步研究的关键问题,以期为今后的应用研究提供理论基础和技术参考。
Studying the diffusion law of harmful chemical gases and building its diffusion model helps to quickly predict the diffusion range of harmful chemical gases under the condition of sudden chemical accidents,thus providing support for taking corresponding rescue actions.In this paper,based on the Physics-Informed Neural Network(PINN)model with physical information constraints and the Navier-Stokes equation of fluid mechanics,a convolutional neural network(CNN)was combined with a Long Short-term Memory(LSTM)network structure.The focus was on the preliminary study of PINN's network structure construction,network algorithm flow,and key issues to be solved in the process of harmful gas prediction,and the basic structure of the PINN based model was designed and built.On this basis,several key issues such as the applicability of N-S equation and PINN model driving mode that need further research were proposed to provide theoretical basis and technical reference for future application research.
作者
田旭光
左钦文
张杰民
张成名
孙茂盛
TIAN Xuguang;ZUO Qinwen;ZHANG Jiemin;ZHANG Chengming;SUN Maosheng(State Key Laboratory of NBC Protection for Civilian,Beijing 102205,China)
出处
《防化研究》
2023年第4期59-65,共7页
CBRN DEFENSE
关键词
有害化学气体
物理信息约束的神经网络
扩散预测
网络构建
harmful chemical gas
Physics-informed Neural Network
diffusion prediction
network construction
作者简介
第一作者:田旭光(1981-),男,博士,助理研究员,主要从事装备仿真研究。E-mail:tiancug@163.com;通信作者:左钦文(1978-),男,博士,副研究员,主要从事装备仿真研究。E-mail:403951765@qq.com。
