[1]王璐瑶,孙秉才,段礼祥,等.面向天然气阀室泄漏扩散的喷流感知时空预测方法*[J].中国安全生产科学技术,2026,22(4):114-120.[doi:10.11731/j.issn.1673-193x.2026.04.014]
WANG Luyao,SUN Bingcai,DUAN Lixiang,et al.A jet-aware spatiotemporal prediction method for leakage diffusion in natural gas valve chambers[J].Journal of Safety Science and Technology,2026,22(4):114-120.[doi:10.11731/j.issn.1673-193x.2026.04.014]
点击复制
面向天然气阀室泄漏扩散的喷流感知时空预测方法*
WANG Luyao,SUN Bingcai,DUAN Lixiang,et al.A jet-aware spatiotemporal prediction method for leakage diffusion in natural gas valve chambers[J].Journal of Safety Science and Technology,2026,22(4):114-120.[doi:10.11731/j.issn.1673-193x.2026.04.014]
《中国安全生产科学技术》[ISSN:1673-193X/CN:11-5335/TB]
- 卷:
- 22
- 期数:
- 2026年4期
- 页码:
- 114-120
- 栏目:
- 安全工程技术
- 出版日期:
- 2026-04-30
文章信息/Info
- Title:
- A jet-aware spatiotemporal prediction method for leakage diffusion in natural gas valve chambers
- 文章编号:
- 1673-193X(2026)-04-0114-07
- Keywords:
- natural gas valve chambers; gas leakage diffusion; jet-aware time-varying graph; time delay alignment; DCRNN
- 分类号:
- X931；TE832
- 文献标志码:
- A
- 摘要:
- 为解决天然气阀室泄漏扩散预测方法中时空关联性建模多依赖静态有向图结构，难以有效建模无向扩散与有向“喷流”的动态交互关系以及气体传播的时间滞后效应问题，提出1种基于序列到序列架构（Seq2Seq）的喷流感知多时滞扩散卷积循环神经网络（Seq2Seq-JET-MDCRNN）预测方法。首先在传统DCRNN静态图结构基础上，引入考虑喷流信息（JET）的阀室监测点动态有向图结构，融合动量主导与扩散主导2种传输机制；然后设计多时滞扩散卷积循环神经网络（MDCRNN），显式建模跨监测点的传播时滞，从而对齐气体“先近后远”的物理到达顺序，实现更符合实际传播过程的时空融合浓度预测；最后采用Fluent软件建立1∶1天然气阀室模型，并以生成的多工况泄漏扩散数据集为例进行验证。研究结果表明：Seq2Seq-JET-MDCRNN在平均绝对误差（MAE）指标上相比Seq2Seq-DCRNN降低13.5%，较传统RNN降低34.2%。研究结果可为天然气阀室泄漏浓度预测预警提供参考。
- Abstract:
- In order to address the limitations of existing spatiotemporal correlation modeling approaches for predicting leakage diffusion in natural gas valve chambers,which predominantly rely on static directed graph structures and struggle to effectively capture the dynamic interaction between undirected diffusion and directed jet flow as well as the temporal lag effects of gas propagation,a jet-aware multi-delay diffusion convolutional recurrent neural network based on a sequence-to-sequence architecture,termed Seq2Seq-JET-MDCRNN,is proposed.Building on the conventional static graph structure of diffusion convolutional recurrent neural networks (DCRNN),a dynamic directed graph structure incorporating jet flow information is introduced for valve chambers monitoring points,integrating both momentum-dominated and diffusion-dominated transport mechanisms.A multi-delay diffusion convolutional recurrent neural network (MDCRNN) is then designed to explicitly model propagation delays across monitoring points,thereby aligning with the physical arrival sequence of gas that reaches proximal points before distal ones and enabling spatiotemporal concentration predictions that more faithfully represent the actual propagation process.A full-scale (1∶1) natural gas valve chambers model is subsequently established using Flunet software,and a multi-condition leakage diffusion dataset generated from this model is used for validation.The results show that Seq2Seq-JET-MDCRNN reduces the mean absolute error (MAE) by 13.5% relative to Seq2Seq-DCRNN and by 34.2% relative to conventional RNN.These findings provide a reference for leakage concentration prediction and early warning in natural gas valve chambers.
参考文献/References:
［1］李新宏,贾明汭,韩子月,等.穿越城市生活区的天然气管道泄漏连锁爆燃后果评估研究［J］.中国安全生产科学技术,2022,18(8):183-188.
LI Xinhong,JIA Mingrui,HAN Ziyue,et al.Study on sequence assessment of leakage and chain deflagration of natural gas pipeline crossing urban living area［J］.Journal of Safety Science and Technology,2022,18(8):183-188.
［2］李新宏,韩子月,耿志远,等.基于实验与仿真的三维水下气体泄漏预测与安全评估研究［J］.中国安全生产科学技术,2020,16(1):50-54.
LI Xinhong,HAN Ziyue,GENG Zhiyuan,et al.Study on three-dimensional prediction and safety assessment of underwater gas leakage based on experiment and simulation［J］.Journal of Safety Science and Technology,2020,16(1):50-54.
［3］张杰,王璐瑶,赵杰,等.储油库VOCs多呼吸阀排放的扩散叠加效应分析［J/OL］.环境工程学报,1-10［2026-03-09］.https://link.cnki.net/urlid/11.5591.X.20251127.0942.008.
［4］QIAN F,CHEN L,LI J,et al.Direct prediction of the toxic gas diffusion rule in a real environment based on LSTM［J］.International Journal of Environmental Research and Public Health,2019,16(12):2133.
［5］赵旭辉.基于深度学习的有害气体扩散预测模型研究及其应用［D］.郑州：郑州大学,2023.
［6］王博乔.FLNG上甲板可燃气体泄漏燃爆事故后果快速预测与泄漏源反演方法研究［D］.大连：大连海事大学,2022.
［7］XU Y H,SHA Y Y,WANG C,et al.Comparative studies of predictive models for unsteady flow fields based on deep learning and proper orthogonal decomposition［J］.Ocean Engineering,2023,272:113935.
［8］WANG X Q,XU N K,MENG X R,et al.Prediction of gas concentration based on LSTM-LightGBM variable weight combination model［J］.Energies,2022,15(3):827.
［9］GAO K,ZHOU Z,QIN Y.Gas concentration prediction by LSTM network combined with wavelet thresholding denoising and phase space reconstruction［J］.Heliyon,2024,10(7):e28112.
［10］张倩倩,李敏,耿绍胜,等.考虑时空耦合关系的电力变压器油中溶解气体多尺度融合预测方法［J］.绝缘材料,2025,58(6):122-130.
ZHANG Qianqian,LI Min,GENG Shaosheng,et al.Multi-scale fusion prediction method of dissolved gas in power transformer oil considering spatio-temporal coupling relationship［J］.Insulating Materials,2025,58(6):122-130.
［11］杜鑫,黄萍,唐崇年,等.基于图时间卷积的变压器油中溶解气体预测方法［J］.广东电力,2025,38(4):101-108.
DU Xin,HUANG Ping,TANG Chongnian,et al.Prediction of dissolved gases in transformer oil based on graph time convolution［J］.Guangdong Electric Power,2025,38(4):101-108.
［12］李紫豪,何怡刚,周亚中,等.计及时空特性的变压器油中溶解气体预测模型［J］.电子测量与仪器学报,2025,39(3):1-12.
LI Zihao,HE Yigang,ZHOU Yazhong,et al.Prediction model for dissolved gases in oil considering spatiotemporal characteristics［J］.Journal of Electronic Measurement and Instrumentation,2025,39(3):1-12.
［13］王昱晨,骆正山,骆济豪.时空融合深度学习模型的有毒重气扩散预测［J］.安全与环境学报,2025,25(5):1911-1919.
WANG Yuchen,LUO Zhengshan,LUO Jihao.Spatio-temporal fusion of a deep learning model for toxic heavy gas dispersion prediction［J］.Journal of Safety and Environment,2025,25(5):1911-1919.
［14］闫佳和,李红辉,孙婧,等.面向交通流预测的时空图神经网络发展综述［J］.计算机工程与应用,2025,61(22):1-19.
YAN Jiahe,LI Honghui,SUN Jing,et al.Review on development of spatio-temporal graph neural networks for traffic flow prediction［J］.Computer Engineering and Applications,2025,61(22):1-19.
［15］MALLICK T,BALAPRAKASH P,RASK E,et al.Graph-partitioning-based diffusion convolutional recurrent neural network for large-scale traffic forecasting［J］.Transportation Research Record Journal of the Transportation Research Board,2020,2674(9):473-488.
［16］宋凯磊,张欣海,侯位昭,等.融合多模态信息的Prophet-DCRNN交通预测方法［J］.中国电子科学研究院学报,2021,16(3):250-254+264.
SONG Kailei,ZHANG Xinhai,HOU Weizhao,et al.Prophet-DCRNN traffic forecasting method with multi-modal information［J］.Journal of China Academy of Electronics and Information Technology,2021,16(3):250-254,264.
［17］WANG Y C,HAN J C,ZHOU Y,et al.Application of novel deep learning model for high-frequency water quality prediction in multi-station river networks［J］.Journal of Environmental Chemical Engineering,2025,13(6):119673.
［18］WANG D S,WANG H W,LU K F,et al.Regional prediction of ozone and fine particulate matter using diffusion convolutional recurrent neural network［J］.International Journal of Environmental Research and Public Health,2022,19(7):3988.
［19］赵朋,王建伟,孙茂棚,等.高速公路突发事件恢复重建期交通量预测［J］.长安大学学报(自然科学版),2018,38(5):139-145.
ZHAO Peng,WANG Jianwei,SUN Maopeng,et al.Traffic flow prediction of expressway traffic emergency recovery and reconstruction period［J］.Journal of Chang’an University (Natural Science Edition),2018,38(5):139-145.
［20］钱浩.天然气泄漏扩散实验与高斯预测模型优化［J］.天然气工业,2024,44(7):154-164.
QIAN Hao.Experiments of natural gas leakage diffusion and optimization of the Gaussian prediction model［J］.Natural Gas Industry,2024,44(7):154-164.
相似文献/References:
备注/Memo
- 备注/Memo:
-
收稿日期： 2025-12-30
* 基金项目： 国家重点研发计划项目（2024YFB3409300）；中国石油天然气集团有限公司科技项目（2023DJ6508）
作者简介： 王璐瑶，硕士研究生，主要研究方向为安全监测与智能诊断。
通信作者： 段礼祥，博士，教授，主要研究方向为安全监测与智能诊断、安全大数据与人工智能。
