基于CFD模拟的人工神经网络动态溯源模型

doi:10.16085/j.issn.1000-6613.2025-0551

化工进展 ›› 2025, Vol. 44 ›› Issue (8): 4772-4784.DOI: 10.16085/j.issn.1000-6613.2025-0551

• 过程系统工程的模拟与仿真 • 上一篇

基于CFD模拟的人工神经网络动态溯源模型

史天乐¹^,²(), 李飞²^,³(), 陈昇⁴(), 卢春喜¹, 王维²^,³

^1.中国石油大学(北京)化学工程与环境学院，北京 102249
^2.中国科学院过程工程研究所介科学与工程国家重点实验室，北京 100190
^3.中国科学院大学化学工程学院，北京 100049
^4.国家市场监督管理总局技术创新中心 (炼油与化工装备风险防控)，中国特种设备检测研究院，北京 100029

收稿日期:2025-04-14 修回日期:2025-06-25 出版日期:2025-08-25 发布日期:2025-09-08
通讯作者: 李飞,陈昇
作者简介:史天乐（2000—），硕士研究生，研究方向为气体泄漏仿真及机器学习。E-mail：tlshi@ipe.ac.cn。
基金资助:
国家重点研发计划(2022YFC3004504);国家重点研发计划(2022YFC3902505);中国科学院先导专项(XDA0490102);国家自然科学基金(22208379);国家自然科学基金(U23A20374);国家自然科学基金(22161142006);国家自然科学基金(51876212)

Dynamic source localization model based on CFD simulated artificial neural network

SHI Tianle¹^,²(), LI Fei²^,³(), CHEN Sheng⁴(), LU Chunxi¹, WANG Wei²^,³

^1.College of Chemical Engineering and Environment, China University of Petroleum, Beijing 102249, China
^2.State Key Laboratory of Mesoscience and Engineering, Institute of Process Engineering, Chinese Academy of Sciences, Beijing 100190, China
^3.School of Chemical Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
^4.Technology Innovation Center of Risk Prevention and Control of Refining and Chemical Equipment，State Administration for Market Regulation，China Special Equipment Inspection and Research Institute, Beijing 100029, China

Received:2025-04-14 Revised:2025-06-25 Online:2025-08-25 Published:2025-09-08
Contact: LI Fei, CHEN Sheng

摘要/Abstract

摘要：

危险气体泄漏事故早期处理不当，可能会引发二次燃爆等次生灾害，因此开发一种快速泄漏源定位的气体溯源方法至关重要。气体溯源是气体扩散的逆问题，在科学研究和工程应用中仍具有挑战性，人工神经网络与溯源定位方案的结合为解决这一反问题提供了一种可行途径，有望实现快速准确的溯源定位。本文基于计算流体动力学模拟结果建立动态气体溯源数据集，搭建了基于传感器数据序列实时预测泄漏源位置的长短期记忆神经网络动态溯源模型，并对模型进行训练和优化。结果表明：基于人工神经网络的动态溯源模型成功实现了对泄漏源的准确预测，预测点与真实泄漏源位置的距离在20m以内，模型的准确率达97.49%。在输入一组序列浓度数据后，可以在0.04737s内预测泄漏源的初步位置，显著快于传统的溯源定位方法。

关键词: 计算流体力学, 模拟, 神经网络, 动态溯源定位模型, 反问题

Abstract:

In the early stages of hazardous gas leakage accidents, improper handling can lead to secondary incidents such as combustion and explosion. Thus it is crucial to develop a gas tracing method for rapid localization of leak source. Gas tracing and localization, an inverse problem of gas diffusion, remains challenging in scientific research and engineering applications. Artificial neural networks, integrated with tracing and localization schemes offer a promising solution for this inverse problem, enabling rapid and accurate source localization. The dynamic source localization datasets are extracted from computational fluid dynamics simulation results. A long short-term memory neural network model for real-time prediction of leakage source locations based on sensor data sequence is established and optimized. Results show that the artificial neural network-based localization model can accurately predict the leakage source, with predicted points within 20m of the actual source location, achieving an accuracy of 97.49%. After a set of sequential concentration data is input, the preliminary location of the leakage source can be predicted within 0.04737s, which is significantly faster than traditional source localization methods.

Key words: CFD, simulation, neural network, dynamic source localization model, inverse problem

中图分类号:

TQ021.1

史天乐, 李飞, 陈昇, 卢春喜, 王维. 基于CFD模拟的人工神经网络动态溯源模型[J]. 化工进展, 2025, 44(8): 4772-4784.

SHI Tianle, LI Fei, CHEN Sheng, LU Chunxi, WANG Wei. Dynamic source localization model based on CFD simulated artificial neural network[J]. Chemical Industry and Engineering Progress, 2025, 44(8): 4772-4784.

图/表 24

图1 RNN网络中的数据传输原理

图2 LSTM结构单元

图3 基于深度学习的气体泄漏溯源模型开发流程

图4 溯源测点的布置方案

表1 CFD模型公式

方程	公式
质量方程	$∂ ρ ∂ t + ∇ ⋅ ρ u = 0$
动量方程	$∂ ∂ t ρ u + ∇ ⋅ ρ u u = - ∇ p + ∇ ⋅ τ + ρ g + F$
	$τ = μ e f f ∇ u + ∇ u T - 23 ∇ ⋅ u I$
组分输运方程	$∂ ρ Y i ∂ t + ∇ ⋅ ρ u Y i = ∇ ⋅ ρ D e f f ∇ Y i + S i$
标准k-ε模型	$∂ ρ k ∂ t + ∇ ⋅ ρ k u = ∇ ⋅ μ + μ t σ k ∇ k + G k + G b - ρ ε - Y m$
	$∂ ρ ε ∂ t + ∇ ⋅ ρ ε u = ∇ ⋅ μ + μ t σ ε ∇ ε + C 1 ε ε k G k + C 3 k G b - C 2 ε ρ ε 2 k$
	$μ t = ρ C μ k 2 ε$
	$D t = μ t ρ S c t$
湍流扩散模型	$μ e f f = μ t + μ$
	$D e f f = D t + D$

表1 CFD模型公式

方程	公式
质量方程	$∂ ρ ∂ t + ∇ ⋅ ρ u = 0$
动量方程	$∂ ∂ t ρ u + ∇ ⋅ ρ u u = - ∇ p + ∇ ⋅ τ + ρ g + F$
	$τ = μ e f f ∇ u + ∇ u T - 23 ∇ ⋅ u I$
组分输运方程	$∂ ρ Y i ∂ t + ∇ ⋅ ρ u Y i = ∇ ⋅ ρ D e f f ∇ Y i + S i$
标准k-ε模型	$∂ ρ k ∂ t + ∇ ⋅ ρ k u = ∇ ⋅ μ + μ t σ k ∇ k + G k + G b - ρ ε - Y m$
	$∂ ρ ε ∂ t + ∇ ⋅ ρ ε u = ∇ ⋅ μ + μ t σ ε ∇ ε + C 1 ε ε k G k + C 3 k G b - C 2 ε ρ ε 2 k$
	$μ t = ρ C μ k 2 ε$
	$D t = μ t ρ S c t$
湍流扩散模型	$μ e f f = μ t + μ$
	$D e f f = D t + D$

图5 三维计算域几何模型构建与网格离散化示意图

图6 Jack Rabbit实验布置

表2 数值模拟关键参数及边界条件设置

模拟参数	模拟设置
网络数	127831
压力出口	1atm
空气-速度入口	3.95m/s
氯气-速度入口	0.08m/s
壁面边界	无滑移
动量离散	二阶迎风格式
湍流黏度比	10
湍流强度	0.05

表3 模拟扩散角度与实验测量结果的比较

项目	下风口200m $θ 200$ /(°)	下风口500m $θ 500$ /(°)
实验值	59.79	44.95
计算值	59.79	35

表3 模拟扩散角度与实验测量结果的比较

项目	下风口200m $θ 200$ /(°)	下风口500m $θ 500$ /(°)
实验值	59.79	44.95
计算值	59.79	35

图7 模拟浓度与实验测量结果对比

图8 LSTM模型的原始数据重塑原理

图9 LSTM网络模型原理

表4 不同LSTM网络模型训练参数量训练结果对比

模型	网络结构（L表示LSTM层）	模型训练参数量	预测准确率/%
Model-1	14-25L-27L-41L-25-30-28-3	24197	48.8
Model-2	14-50L-44L-41L-50-35-28-3	49344	50.1
Model-3	14-75L-48L-41L-75-48-28-3	74481	52.1
Model-4	14-84L-62L-41L-100-60-28-3	99579	53.0
Model-5	14-100L-80L-41L-110-80-28-3	140667	54.4
Model-6	14-140L-90L-41L-100-80-28-3	207327	55.1

图10 不同LSTM模型训练参数量训练效果对比

图11 三种不同LSTM网络模型输入方式对比

图12 不同模型输入方式训练效果对比

图13 基于不同数据量的训练效果对比

图14 基于时间窗滑动的LSTM输入序列划分策略与重叠机制示意图

图15 输入序列长度对溯源模型训练损失及预测精度的影响

图16 Dropout随机神经元失活机制原理

图17 Dropout概率对训练性能的影响

图18 多距离阈值下泄漏源定位模型的灵敏度分析

图19 随机泄漏事故溯源预测点与真实泄漏点对比

图20 X、Y的真实点与预测点对比

参考文献 32

[1]	杨华磊, 杨敏. 碳达峰碳中和：中国式现代化的能源转型之路[J]. 经济问题, 2024,(3): 1-7.
	YANG Hualei, YANG Min. Peak carbon emission and carbon neutrality: China’s path to energy transition in modernization[J]. On Economic Problems, 2024,(3): 1-7.
[2]	程硕, 杨富强. 2011—2020年我国危险化学品事故统计及灰色关联分析[J]. 应用化工, 2023, 52(1): 193-198.
	CHENG Shuo, YANG Fuqiang. Statistical and grey relation analysis of hazardous chemicals accidents in China from 2011 to 2020[J]. Applied Chemical Industy, 2023, 52(1): 193-198.
[3]	YANG Dongdong, CHEN Guoming, DAI Ziliang. Accident modeling of toxic gas-containing flammable gas release and explosion on an offshore platform[J]. Journal of Loss Prevention in the Process Industries, 2020, 65: 104118.
[4]	MACDONALD Chris, YANG Michael, LEARN Shawn, et al. Transferable pipeline rupture detection using multiple artificial intelligence classifiers during transient operations[J]. Journal of Pressure Vessel Technology, 2022, 144(4): 041802.
[5]	DESAI Dhwani, MEHENDALE Ninad. A review on sound source localization systems[J]. Archives of Computational Methods in Engineering, 2022, 29(7): 4631-4642.
[6]	CHEN J C, YAO K, HUDSON R E. Source localization and beamforming[J]. Signal Processing Magazine IEEE, 2002, 19(2): 30-39.
[7]	LOUTFI Amy, CORADESCHI Silvia. Odor recognition for intelligent systems[J]. IEEE intelligent Systems, 2008, 23(1): 41-48.
[8]	CHO Jaehoon, KIM Hyunseung, GEBRESELASSIE Addis Lulu, et al. Deep neural network and random forest classifier for source tracking of chemical leaks using fence monitoring data[J]. Journal of Loss Prevention in the Process Industries, 2018, 56: 548-558.
[9]	KIM Hyunseung, GEBRESELASSIE Addis Lulu, DAN Seungkyu, et al. Random forest classifier for real-time chemical leak source tracking using fence-monitoring sensors[J]. Korean Journal of Chemical Engineering, 2018, 35(6): 1231-1239.
[10]	FRESNO José, ROBLES Guillermo, Juan MARTÍNEZ-TARIFA, et al. Survey on the performance of source localization algorithms[J]. Sensors, 2017, 17(11): 2666.
[11]	Hing Cheung SO. Source localization: Algorithms and analysis[M]. Hoboken, NJ: Wiley, 2011, 25-66.
[12]	GOLSTON Levi M, AUBUT Nicholas F, FRISH Michael B, et al. Natural gas fugitive leak detection using an unmanned aerial vehicle: Localization and quantification of emission rate[J]. Atmosphere, 2018, 9(9): 333.
[13]	Andrew RUSSELL R. Robotic location of underground chemical sources[J]. Robotica, 2004, 22(1): 109-115.
[14]	LI Wei, FARRELL Jay A, CARD Ring T. Tracking of fluid-advected odor plumes: Strategies inspired by insect orientation to pheromone[J]. Adaptive Behavior, 2001, 9(3-4): 143-170.
[15]	LOCHMATTER Thomas, RAEMY Xavier, MATTHEY Loic, et al. A comparison of casting and spiraling algorithms for odor source localization in laminar flow[C]//2008 IEEE International Conference on Robotics and Automation. Pasadena, CA, USA： IEEE, 2008: 1138-1143.
[16]	LI Zhiqiang, Rizzo DONNA M, NANCY Hayden. Utilizing artificial neural networks to backtrack source location[C]//Proceedings of the 3rd International Congress on Environmental Modelling and Software. Burlington, Vermont, USA, 2006.
[17]	HUANG Zhaoqiong, XU Ji, GONG Zaixiao, WANG Haibin, et al. Source localization using deep neural networks in a shallow water environment[J]. The Journal of the Acoustical Society of America, 2018, 143(5): 2922-2932.
[18]	KIM Hyunseung, PARK Myeongnam, KIM Chang Won, et al. Source localization for hazardous material release in an outdoor chemical plant via a combination of LSTM-RNN and CFD simulation[J]. Computers & Chemical Engineering, 2019, 125(9): 476-489.
[19]	ASEA Brown Boveri. ABB参加中关村论坛, 共话绿色智能未来[EB/OL]. 瑞士: ABB集团， 2023. (2023-05-31)..
	ASEA Brown Boveri. ABB participates in Zhongguancun Forum to discuss a greener and smarter future[EB/OL]. Switzerland: ABB Group, 2023. (2023-05-31). .
[20]	王凯艺. 移动“5G”赋能制造!宁波再增5个省级未来工厂试点[EB/OL]. 杭州:浙江日报, 2023. (2023-08-01)[2025-06-25]. .
	WANG Kaiyi. Mobile “5G” empowers manufacturing! Ningbo adds 5 more provincial-level future factory pilots[EB/OL]. Hangzhou: Zhejiang Daily, 2023. (2023-08-01)[2025-06-25]..
[21]	SUTTON Oliver George. The theoretical distribution of airborne pollution from factory chimneys[J]. Quarterly Journal of the Royal Meteorological Society, 1947, 73(317-318): 426-436.
[22]	REN Guanlong, SUN Haijun, Chen Fangjun, et al. CFD investigation of structural effects of internal gas intake on powder conveying performance in fuel supply systems for aerospace engines[J]. Particuology, 2024, 92: 140-154.
[23]	REN Guanlong, SUN Haijun, XU Yihua, et al. Effect of Intake position on powder fluidization and conveying characteristics in powder supply device[J]. Chemical Engineering and Processing—Process Intensification, 2023, 183: 109240.
[24]	HOCHREITER Sepp, Jürgen SCHMIDHUBER. Long short-term memory[J]. Neural Computation, 1997 9(8): 1735-1780.
[25]	马巍巍, 徐鸣, 刘香芝, 等. Aspen Plus模拟软件与Aloha软件在化工安全设计中的教学应用——基于CDIO模式[J]. 化学教育(中英文), 2025, 46(2): 117-122.
	MA Weiwei, XU Ming, LIU Xiangzhi, et al. Teaching application of Aspen Plus simulation software and Aloha software in chemical safety design based on CDIO model[J]. Chinese Journal of Chemical Education, 2025, 46(2): 117-122.
[26]	GANT Simon, WEIL Jeffrey, Luca Monache, et al. Dense gas dispersion model development and testing for the Jack Rabbit Ⅱ phase 1 chlorine release experiments[J]. Atmospheric Environment, 2018, 192: 218-240.
[27]	GRAVES Alex, MOHAMED Abdel-rahman, HINTON Geoffrey. Speech recognition with deep recurrent neural networks[C]//2013 IEEE International Conference on Acoustics, Speech, and Signal Processing. 2013. Vancouver, BC, Canada: IEEE, 2013: 6645-6649.
[28]	HAMMERLA Young Ngo, HALLORAN Sean, PLOETZ Torsten. Deep, convolutional, and recurrent models for human activity recognition using wearables[J]. CoRR, 2016, abs/1604.08880.
[29]	中华人民共和国工业和信息化部. 液氯泄漏的处理处置方法: [S]. 北京:化学工业出版社有限公司, 2014.
	Ministry of Industry and Information of the People’s Republic of China. Treatment and disposal method for liquid chlorine spill: [S]. Beijing: Chemical Industry Press Co., Ltd., 2014.
[30]	郝腾腾, 郑欣, 王慧宇. 基于ALOHA后果模拟的液氨泄漏事故应急策略研究[J]. 安全, 2021, 42(6): 35-40.
	HAO Tengteng, ZHENG Xin, WANG Huiyu. Research on emergency strategy of liquid ammonia leakage accident based on ALOHA consequence simulation[J]. Safety & Security, 2021, 42(6): 35-40.
[31]	张锦荣. 道路运输危险气体泄漏扩散模型构建与应急管理研究[D]. 西安: 长安大学, 2016.
	ZHANG Jinrong. Research on model construction of hazardous gas diffusion after leakage in road transportation and emergency management[D]. Xi’an: Chang’an University, 2016.
[32]	廖倩雯. 天然气瞬时泄漏扩散及燃爆区域划分研究[D]. 大连: 大连交通大学, 2014.
	LIAO Qianwen. Study on the diffusion of the instantaneous leakage of natural gas and its blasting area[D]. Dalian: Dalian Jiaotong University, 2014.

基于CFD模拟的人工神经网络动态溯源模型

Dynamic source localization model based on CFD simulated artificial neural network

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 24

参考文献 32

相关文章 15

编辑推荐

Metrics

本文评价

[1]	刘燕燕, 李飞泉, 刘栋, 王俊涛, 罗雪. 纳观尺度下再生沥青-集料界面性质的分子模拟[J]. 化工进展, 2025, 44(8): 4302-4310.
[2]	王肖肖, 孔福林, 李小宇, 任永强, 许世森. CO₂吸收剂在开孔规整填料表面微尺度流动模拟[J]. 化工进展, 2025, 44(8): 4311-4321.
[3]	李艳平, 杨涛, 王洪勋, 张城, 温国胜, 韩治成, 蓝公家, 严大洲. 三氯氢硅在氢气氛中的热分解及还原体系的反应分子动力学模拟[J]. 化工进展, 2025, 44(8): 4322-4330.
[4]	刘力涵, 王琪君, 王轩, 彭阳峰, 徐小飞. 丁苯橡胶应力软化的全原子分子动力学模拟[J]. 化工进展, 2025, 44(8): 4331-4340.
[5]	齐妍, 常昊, 张磊. 基于分子动力学模拟的结构性产品配方设计方法[J]. 化工进展, 2025, 44(8): 4341-4351.
[6]	黄可儿, 刘佳豪, 李昊明, 周天航, 高金森, 蓝兴英. 基于分子动力学模拟的胺溶剂碳捕集过程自扩散系数[J]. 化工进展, 2025, 44(8): 4352-4364.
[7]	李卡, 夏宇轩, 吴晓琴, 易兰, 罗浩. 双层多孔介质燃烧反应器的孔隙尺度计算流体动力学模拟[J]. 化工进展, 2025, 44(8): 4381-4393.
[8]	戴贵龙, 王孝宇, 皇甫江飞, 龚凌诸. 孔隙尺度下Laguerre Voronoi开孔泡沫的对流传热特性[J]. 化工进展, 2025, 44(8): 4394-4407.
[9]	沈宪琨, 贾志勇, 蓝晓程, 王铁峰. CFD-PBM耦合模型用于浆态床反应器的研究进展[J]. 化工进展, 2025, 44(8): 4408-4418.
[10]	朱孝忠, 房韡, 赵毅. 基于深度VGG模型在乙烯裂解炉装置的预测应用[J]. 化工进展, 2025, 44(8): 4419-4429.
[11]	李增, 赵云鹏, 李宇慧, 柳楠, 朱春梦, 石孝刚, 高金森, 蓝兴英. 基于CFD模拟的催化裂化沉降器跑剂异常诊断[J]. 化工进展, 2025, 44(8): 4430-4442.
[12]	王兆霖, 张志刚, 周静, 高琛, 彭克臣, 姜敏迪, 奚溪, 徐胜利, 刘红. Gyroid三周期极小曲面换热构件流动换热特性[J]. 化工进展, 2025, 44(8): 4454-4462.
[13]	续文钧, 张建波, 郭彦霞, 李会泉, 李少鹏, 任艺凌. 锚框式桨结构参数对煤气化渣活化过程流场特性的影响[J]. 化工进展, 2025, 44(8): 4463-4477.
[14]	胡佳志, 蒋新宇, 李凡, 李志辉. 临近空间飞行器再入DSMC方法表面催化反应模型[J]. 化工进展, 2025, 44(8): 4478-4487.
[15]	张建伟, 阴苗苗, 董鑫, 冯颖. 基于振荡射流的撞击流反应器混合特性数值模拟[J]. 化工进展, 2025, 44(8): 4488-4499.