基于门控残差跳连异构图神经网络的天然气管网动态仿真方法

Transient simulation method for natural gas pipeline networks based on gated residual jump-knowledge heterogeneous graph neural networks

摘要

数据驱动方法有望突破管网机制仿真模型的适用规模与效率瓶颈,但广泛采用的图神经网络难以描述异构设备机制及流动参数长距离稳定传播。 为此,将管网抽象为含多类型节点与边的异构图,构建物理引导的各向异性消息及异构关系门控聚合机制,增强模型对不同设备水力特性差异的表达能力;基于初始特征残差连接,引入层内门控更新及自适应跳层融合机制,实现跨尺度融合并提升模型传播稳定性与缓解分支的过平滑问题;以此构建了基于门控残差跳连异构图神经网络的管网系统动态仿真方法。 结果表明:模型对典型异构管网和长链短分支管网仿真的压力平均相对偏差保持一致;某省级管网实时动态仿真精度与机制模型的平均相对偏差分别为 3. 55%和 3. 13%,仿真速度提升超 60%。

Abstract

Data-driven methods are expected to overcome the scalability and efficiency bottlenecks of mechanism-based pipe network simulation models. However, widely used graph neural networks struggle to describe the mechanisms of heterogeneous devices and the long-distance stable propagation of flow parameters. To address this problem, the pipe network was abstracted as a heterogeneous graph containing multiple types of nodes and edges. A physically guided anisotropic messaging and heterogeneous relation gated aggregation mechanism was constructed to enhance the model̍s ability to represent hydraulic characteristic differences across devices. Based on initial feature residual connections, an intra-layer gated update and an adaptive jumping knowledge fusion mechanism were introduced to achieve cross-scale fusion, improve propagation stability in deep models, and mitigate over-smoothing in branches. Thereby, a dynamic simulation method for pipe network systems based on a gated residual jump knowledge heterogeneous graph neural network was developed. Results show that the proposed model achieves accuracy comparable to the mechanistic model for typical heterogeneous networks and long-chain short-branch networks. For real-time dynamic simulation of a provincial-level network, the model achieves mean relative pressure deviations of 3. 55% and 3. 13% respectively, while improving simulation speed by over 60%.

关键词

天然气管网 / 动态仿真 / 图神经网络 / 深度学习

Key words

gas pipeline network / transient simulation / graph neural networks / deep learning

参考文献

[1] 李长俊, 宇波, 张对红, 等 . 油气综合立体调运关键技术现状与趋势[J]. 前瞻科技, 2024, 3(2): 39-49.

[2] LI Changjun, YU Bo, ZHANG Duihong, et al. Current situation and trends of key technologies for integrated multidimensional oil and gas transportation[J]. Science and Technology Foresinght, 2024, 3(2): 39-49.

[3] 季静. 全球天然氢成因机制及中国天然氢前景[J]. 断块油气田, 2025, 32(6): 976-983, 994.

[4] JI Jing. Genetic mechanisms of global natural hydrogen and prospects of natural hydrogen in China[J]. Fault-Block Oil & Gas Field, 2025, 32(6): 976-983, 994.

[5] 黄维和, 宫敬. 天然气管道与管网多能融合技术展望[J]. 油气储运, 2023, 42(12): 1321-1328.

[6] HUANG Weihe, GONG Jing. Prospect for the development of natural gas network and the multi-energy integration technology in pipeline networks[J]. Oil & Gas Storage and Transportation, 2023, 42(12): 1321-1328.

[7] 宫敬, 史博会, 李晓平, 等 . 油气管网仿真技术在智能管网建设中的应用及展望[J]. 油气储运, 2023, 42(9): 988-997.

[8] GONG Jing, SHI Bohui, LI Xiaoping, et al. Application of oil and gas pipeline network simulation technologies in construction of intelligent pipeline network and its prospect[J]. Oil & Gas Storage and Transportation, 2023, 42(9): 988-997.

[9] 张对红, 杨毅. 大型复杂天然气管网在线仿真软件国产化研发及应用[J]. 油气储运, 2025, 44 (10): 1157-1165.

[10] ZHANG Duihong, YANG Yi. Domestic R & D and application of online simulation software for large-scale and complex natural gas pipeline networks[J]. Oil & Gas Storage and Transportation, 2025, 44(10): 1157-1165.

[11] WYLIE E B, STREETER V L, STONER M A . Unsteady-state natural-gas calculations in complex pipe systems[J]. Society of Petroleum Engineers Journal, 1974, 14(1): 35-43.

[12] PAMBOUR K A, BOLADO-LAVIN R, DIJKEMA G P J . An integrated transient model for simulating the operation of natural gas transport systems[J]. Journal of Natural Gas Science and Engineering, 2016, 28: 672-690.

[13] 王秀芳, 刘源, 李月明. 基于动态深度可分离卷积神经网络的管道泄漏孔径识别[J]. 中国石油大学学报(自然科学版), 2024, 48(5): 183-189.

[14] WANG Xiufang, LIU Yuan, LI Yueming. Identification of pipe leak apertures based on dynamic depth-separable convolutional neural networks[J]. Journal of China University of Petroleum (Edition of Natural Science), 2024, 48(5): 183-189.

[15] 邵必林, 任萌, 田宁. 基于 STL-XGBoost-NBEATSx 的小时天然气负荷预测[J]. 中国石油大学学报(自然科学版), 2024, 48(3): 170-179.

[16] SHAO Bilin, REN Meng, TIAN Ning. Hourly natural gas load forecast based on STL-XGBoost-NBEATSx[J]. Journal of China University of Petroleum (Edition of Natural Science), 2024, 48(3): 170-179.

[17] 郭海伟. 基于增量学习的储气库井底压力快速计算方法[J]. 断块油气田, 2025, 32(2): 292-299.

[18] GUO Haiwei. Rapid calculation method of bottom hole pressure for gas storage based on incremental learning[J]. Fault-Block Oil and Gas Field, 2025, 32(2): 292-299.

[19] ZHOU D, JIA X, MA S, et al. Dynamic simulation of natural gas pipeline network based on interpretable machine learning model[J]. Energy, 2022, 253: 124068.

[20] 赵康, 陈民锋, 王艺文, 等 . TY 含水页岩气藏产量递减规律及可采储量预测[J]. 断块油气田, 2024, 31(3): 395-402.

[21] ZHAO Kang, CHEN Minfeng, WANG Yiwen, et al. Production decline rule and recoverable reserves prediction of TY water-bearing shale gas reservoir[J]. Fault-Block Oil and Gas Field, 2024, 31(3): 395-402.

[22] 金凤, 郭祎, 张岩, 等 . 基于硬约束和物理信息神经网络结合的天然气管网仿真新方法[J]. 天然气工业, 2025, 45(1): 164-174.

[23] JIN Feng, GUO Yi, ZHANG Yan, et al. A new gas pipeline network simulation method based on BHC-PINN[J]. Natural Gas Industry, 2025, 45 (1): 164-174.

[24] 杨毅, 刁洪涛, 陈洁, 等 . 基于 Newton-GNN 的天然气管道系统稳态仿真方法[J]. 石油与天然气化工, 2026, 55(3): 164-174.

[25] YANG Yi, DIAO Hongtao, CHEN Jie, et al. Steady-state simulation method for natural gas pipeline systems based on Newton-GNN[J]. Chemical Engineering of Oil & Gas, 2026, 55(3): 164-174.

[26] SHUKLA K, XU M, TRASK N, et al. Scalable algorithms for physics-informed neural and graph networks[J]. Data-Centric Engineering, 2022, 3: e24.

[27] PENG J Z, HUA Y, AUBRY N, et al. Data and physics-driven modeling for fluid flow with a physics-informed graph convolutional neural network[J]. Ocean Engineering, 2024, 301: 117551.

[28] 李长俊, 张员瑞, 贾文龙, 等 . 大型天然气管网系统在线仿真方法及软件开发[J]. 油气储运, 2022, 41(6): 723-731.

[29] LI Changjun, ZHANG Yuanrui, JIA Wenlong, et al. Online simulation method and software development of large-scale natural gas pipeline network system[J]. Oil & Gas Storage and Transportation, 2022, 41 (6): 723-731.

[30] WU Z, PAN S, CHEN F, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(1): 4-24.

[31] 宇波, 焦开拓, 陈宇杰, 等 . 有限容积法离散方程截断误差本质及其来源[J]. 科学通报, 2023, 68 (10): 1266-1280.

[32] YU Bo, JIAO Kaituo, CHEN Yujie, et al. Essence and composition of truncation errors of discrete equations based on the finite volume method[J]. Chinese Science Bulletin, 2023, 68(10): 1266-1280.

[33] 赵佩尧, 李正烁. 基于离散化和线性化的输气管网动态偏微分方程模型求解方法[J]. 工程科学与技术, 2025, 57(2): 277-288.

[34] ZHAO Peiyao, LI Zhengshuo. Solution for solving dynamic partial differential equation model of natural gas pipeline network based on discretization and linearization[J]. Advanced Engineering Sciences, 2025, 57 (2): 277-288.

[35] SU H, ZIO E, ZHANG J, et al. A systematic hybrid method for real-time prediction of system conditions in natural gas pipeline networks[J]. Journal of Natural Gas Science and Engineering, 2018, 57: 31-44.

[36] RAISSI M, PERDIKARIS P, KARNIADAKIS G E . Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707.

[37] WANG S, TENG Y, PERDIKARIS P . Understanding and mitigating gradient flow pathologies in physics-informed neural networks[J]. SIAM Journal on Scientific Computing, 2021, 43(5): A3055-A3081.

[38] WANG S, YU X, PERDIKARIS P . When and why PINNs fail to train: a neural tangent kernel perspective[J]. Journal of Computational Physics, 2022, 449: 110768.

[39] CUOMO S, DI COLA V S, GIAMPAOLO F, et al. Scientific machine learning through physics-informed neural networks: where we are and what̍s next[J]. Journal of Scientific Computing, 2022, 92(3): 88.

[40] ZHANG C, SHAFIEEZADEH A . Nested physics-informed neural network for analysis of transient flows in natural gas pipelines[J]. Engineering Applications of Artificial Intelligence, 2023, 122: 106073.

[41] SIMA W, WANG J, YANG M, et al. Physics-informed multi-scale adaptive graph learning for unified event detection, localization, and classification in power system[J]. IEEE Transactions on Industrial Informatics, 2026, PP(99): 1-12.

基金资助

国家自然科学基金项目(52274065)

四川省科技计划项目(2024NSFSC1959)