基于物理信息神经网络的泄漏模拟及风险分级  

Leakage Simulation and Risk Classification Based on Physics-Informed Neural Network

作　　者：曾令旗 何睿 侯庆志[2] 毛邓添 王磊 王晓静[3] ZENG Ling-qi;HE Rui;HOU Qing-zhi;MAO Deng-tian;WANG Lei;WANG Xiao-jing(Engineering and Technology Institute of the Safety and Environmental Protection Branch of CNOOC Energy Development Co.,Ltd.;State Key Laboratory of Hydraulic Engineering Intelligent Construction and Operation;School of Chemical Engineering,Tianjin University)

机构地区：[1]中海油能源发展股份有限公司安全环保分公司安全环保工程技术研究院 [2]天津大学水利工程智能建造与运维全国重点实验室 [3]天津大学化工学院

出　　处：《化工自动化及仪表》2025年第2期250-258,300,共10页Control and Instruments in Chemical Industry

基　　金：国家重点研发计划(批准号:2023YFC3049403)资助的课题。

摘　　要：当海上石油平台发生原油或有毒有害气体泄漏时,需要对泄漏的物质进行快速模拟,并且根据模拟结果划分风险等级区域,用来疏散工作人员和抢险维修。针对对流-扩散过程的快速模拟与风险区域划分问题,在改进物理信息神经网络(PINN)方法的基础上,实现了高效率、高精度的快速模拟。对于PINN方法在求解不同参数对流-扩散方程时需要从头训练的问题,提出了一种基于模型参数迁移的训练方案。通过二维污染物对流-扩散数值模拟实验,验证了参数迁移训练方案应对对流方向和速度两类不同变化的有效性。研究结果表明:无论改变对流速度还是对流方向,PINN方法进行模型参数迁移均能有效地提升训练效率,并在一定程度上减小预测误差。最后利用参数迁移训练方案,快速地绘制了污染物对流-扩散风险等级划分图。As for the crude oil or toxic and harmful gas leaked from offshore oil platform,quickly simulating the substances leaked and dividing their risk areas,as well as evacuating operators and carrying out emergency maintenance become necessary.Aiming at fast simulation and risk region division of the convection-diffusion process,having the improved physics-informed neural network(PINN)method based to realize fast simulation with high efficiency and accuracy was implemented.Considering the fact that PINN methods ask for the train from scratch while solving convection-diffusion equations with different parameters,a training scheme based on model parameter transfer was proposed.Through simulation and experiments on a 2D pollutants’convection-diffusion,the parameter transfer training scheme’s effectiveness in dealing with two different changes in convection direction and velocity was verified to show that,the PINN method for model parameter transfer can effectively improve the training efficiency and reduce the prediction error to a certain extent regardless of convection speed or direction changed.Finally,having the parameter transfer training scheme adopted to quickly draw the diagram of pollutants’convection-diffusion risk classification was carried out.

关 键 词：PINN 对流-扩散 参数迁移 风险划分 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]