en
×

分享给微信好友或者朋友圈

使用微信“扫一扫”功能。
作者简介:

樊冬艳(1985-),女,副教授,博士,硕士生导师,研究方向为非常规油气试井及动态分析方法。E-mail: fandongyan2010@126.com。

通信作者:

孙海(1984-),男,教授,博士,博士生导师,研究方向为非常规油气渗流理论及流动机制。E-mail: sunhai@upc.edu.cn。

中图分类号:TE312

文献标识码:A

文章编号:1673-5005(2025)06-0141-11

DOI:10.3969/j.issn.1673-5005.2025.06.014

参考文献 1
张君峰,周志,宋腾,等.中美页岩气勘探开发历程、地质特征和开发利用条件对比及启示[J].石油学报,2022,43(12):1687-1701.ZHANG Junfeng,ZHOU Zhi,SONG Teng,et al.Comparison of exploration and development history,geological characteristics and exploitation conditions of shale gas in China and the United States and its enlightenment[J].Acta Petrolei Sinica,2022,43(12):1687-1701.
参考文献 2
赵康,陈民锋,王艺文,等.四川盆地SZ页岩气藏气井产量变化规律及递减预测新模型[J].中国海上油气,2024,36(3):129-136.ZHAO Kang,CHEN Minfeng,WANG Yiwen,et al.A new model for production variation patterns and decline prediction of SZ shale gas reservoir in Sichuan Basin[J].China Offshore Oil and Gas,2024,36(3):129-136.
参考文献 3
王睿,薛龙龙,党东琪,等.产量递减方程通式的建立及应用[J].石油学报,2023,44(10):1693-1705,1738.WANG Rui,XUE Longlong,DANG Dongqi,et al.Establishment and application of general equation for production decline[J].Acta Petrolei Sinica,2023,44(10):1693-1705,1738.
参考文献 4
ARPS J J.Analysis of decline curves[J].Transactions of the AIME,2016,160:228-247.
参考文献 5
徐兵祥,李相方,HAGHIGHI Manouchehr,等.裂缝性页岩气藏水平井产能预测模型[J].开云电竞投注学报(自然科学版),2013,37(6):92-99,105.XU Bingxiang,LI Xiangfang,HAGHIGHI Manouchehr,et al.A new model for production analysis in naturally fractured shale gas reservoirs[J].Journal of China University of Petroleum(Edition of Natural Science),2013,37(6):92-99,105.
参考文献 6
陈元千,王鑫,刘洋.Blasingame气井变产量拟稳态压降方程存在的问题及其推导与应用[J].中国海上油气,2023,35(6):70-77.CHEN Yuanqian,WANG Xin,LIU Yang.Existing problems of Blasingames pseudo-steady-state pressure drawdown equation with variable rate for gas wells along with derivation and application[J].China Offshore Oil and Gas,2023,35(6):70-77.
参考文献 7
张黎明,陈昕晟,李国欣,等.基于集合和神经网络架构搜索的自动历史拟合方法[J].开云电竞投注学报(自然科学版),2022,46(2):127-136.ZHANG Liming,CHEN Xinsheng,LI Guoxin,et al.An automatic history matching method based on ensemble and neural architecture search[J].Journal of China University of Petroleum(Edition of Natural Science),2022,46(2):127-136.
参考文献 8
CLARKSON C R R,NOBAKHT M,KAVIANI D,et al.Production analysis of tight-gas and shale-gas reservoirs using the dynamic-slippage concept[J].SPE Journal,2012,17(1):230-242.
参考文献 9
林伯韬,郭建成.人工智能在石油工业中的应用现状探讨[J].石油科学通报,2019,4(4):403-413.LIN Botao,GUO Jiancheng.Discussion on current application of artificial intelligence in petroleum industry[J].Petroleum Science Bulletin,2019,4(4):403-413.
参考文献 10
杨磊,王化增,陈子凌.基于支持向量机的油气储量价值等级评价[J].开云电竞投注学报(自然科学版),2012,36(3):192-196.YANG Lei,WANG Huazeng,CHEN Ziling.Assessing value classification of oil and gas reserve based on support vector machine[J].Journal of China University of Petroleum(Edition of Natural Science),2012,36(3):192-196.
参考文献 11
周彩兰,刘敏.BP神经网络在石油产量预测中的应用[J].武汉理工大学学报,2009,31(3):125-129.ZHOU Cailan,LIU Min.Application research on oil production forecasting based on BP neural network[J].Journal of Wuhan University of Technology,2009,31(3):125-129.
参考文献 12
高亚军,唐力辉,王振鹏,等.基于循环神经网络和数据差分处理的油田产量预测方法[J].中国海上油气,2023,35(3):126-136.GAO Yajun,TANG Lihui,WANG Zhenpeng,et al.A method for oilfield production prediction based on recurrent neural network and data differential processing[J].China Offshore Oil and Gas,2023,35(3):126-136.
参考文献 13
刘合,李艳春,杜庆龙,等.基于多变量时间序列模型的高含水期产量预测方法[J].开云电竞投注学报(自然科学版),2023,47(5):103-114.LIU He,LI Yanchun,DU Qinglong,et al.Prediction of production during high water-cut period based on multivariate time series model[J].Journal of China University of Petroleum(Edition of Natural Science),2023,47(5):103-114.
参考文献 14
LI X,MA X,XIAO F,et al.Application of gated recurrent unit(GRU)neural network for smart batch production prediction[J].Energies,2020,13(22):6121.
参考文献 15
樊冬艳,杨灿,孙海,等.基于时间序列相似性与机器学习方法的页岩气井产量预测[J].开云电竞投注学报(自然科学版),2024,48(3):119-126.FAN Dongyan,YANG Can,SUN Hai,et al.Shale gas well production forecasting based on time sequence similarity and machine learning methods[J].Journal of China University of Petroleum(Edition of Natural Science),2024,48(3):119-126.
参考文献 16
ZHA W,LIU Y,WAN Y,et al.Forecasting monthly gas field production based on the CNN-LSTM model[J].Energy,2022,260:124889.
参考文献 17
姜宝胜,白玉湖,徐兵祥,等.基于集成学习的致密气藏产能预测新方法[J].中国海上油气,2024,36(5):120-127.JIANG Baosheng,BAI Yuhu,XU Bingxiang,et al.A novel approach for predicting production capacity of tight gas reservoirs based on ensemble learning[J].China Offshore Oil and Gas,2024,36(5):120-127.
参考文献 18
FAN D,SUN H,YAO J,et al.Well production forecasting based on ARIMA-LSTM model considering manual operations[J].Energy,2021,220:119708.
参考文献 19
GAO M,WEI C,ZHAO X,et al.Production forecasting based on attribute-augmented spatiotemporal graph convolutional network for a typical carbonate reservoir in the middle east[J].Energies,2023,16(1):407.
参考文献 20
DU E,LIU Y,CHENG Z,et al.Production forecasting with the interwell interference by integrating graph convolutional and long short-term memory neural network[J].SPE Reservoir Evaluation & Engineering,2022,25(2):197-213.
参考文献 21
DU E,LIU Y,XUE L,et al.A data-driven model for production prediction of strongly heterogeneous reservoir under uncertainty[J].Geoenergy Science and Engineering,2023,223:211542.
参考文献 22
郭旭洋,金衍,黄雷,等.页岩油气藏水平井井间干扰研究现状和讨论[J].石油钻采工艺,2021,43(3):348-367.GUO Xuyang,JIN Yan,HUANG Lei,et al.Research status and discussion of horizontal well interference in shale oil and gas reservoirs[J].Oil Drilling & Production Technology,2021,43(3):348-367.
参考文献 23
AL-SHAMI T M,JUFAR S R,KUMAR S,et al.A comprehensive review of interwell interference in shale reservoirs[J].Earth-Science Reviews,2023,237:104327.
参考文献 24
程时清,李猛,何佑伟,等.一种用于多级压裂水平井判别来水方向的多井干扰压力分析方法[J].开云电竞投注学报(自然科学版),2018,42(5):81-88.CHENG Shiqing,LI Meng,HE Youwei,et al.A multi-well interference pressure transient analysis method to determine the water source orientation of multi-fractured horizontal well[J].Journal of China University of Petroleum(Edition of Natural Science),2018,42(5):81-88.
参考文献 25
PATEL H,CADWALLADER S,WAMPLER J.Zipper fracturing:taking theory to reality in the eagle ford shale[R].SPE 2445923-MS 2016.
参考文献 26
鹿克峰,苏畅,程超逸.一种评价封闭气藏采气指数、动储量及井间连通性的简便方法[J].中国海上油气,2023,35(4):86-93.LU Kefeng,SU Chang,CHENG Chaoyi.A simple method for evaluating gas production index,dynamic reserves and interwell connectivity of closed gas reservoirs[J].China Offshore Oil and Gas,2023,35(4):86-93.
参考文献 27
LAWAL H,ABOLO N,JACKSON G,et al.A quantitative approach to analyze fracture area loss in shale gas reservoirs[R].SPE 169406-MS,2014.
参考文献 28
ZHAO L,SONG Y,ZHANG C,et al.T-GCN:a temporal graph convolutional network for traffic prediction[J].IEEE Transactions on Intelligent Transportation Systems,2019,21(9):3848-3858.
参考文献 29
BRUNA J,ZAREMBA W,SZLAM A,et al.Spectral networks and locally connected networks on graphs[J].Computer Science,2013,arXiv:1312.6203.
参考文献 30
CHO K,VAN MERRIENBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation [J].Computer Science,2014,arXiv:1406.1078
参考文献 31
杨丽,吴雨茜,王俊丽,等.循环神经网络研究综述[J].计算机应用,2018,38(增刊2):1-6,26.YANG Li,WU Yuqian,WANG Junli,et al.Research on recurrent neural network[J].Journal of Computer Applications,2018,38(sup2):1-6,26.
参考文献 32
HARARY F.The determinant of the adjacency matrix of a graph[J].SIAM Review,1962,4(3):202-210.
目录contents

    摘要

    为了更准确、高效地预测页岩气井产量,针对页岩气藏多井立体井网结合多段压裂水平井开采现状,在考虑人工裂缝展布及多井干扰的基础上,提出一种新的IMWs(interference of multiple wells)-GCN(graph convolutional neural network)-GRU(gate recurrent unit)耦合的页岩气井产量预测机器学习方法。首先,在综合考虑压裂水平井空间位置、裂缝展布和储层渗透率等因素的基础上,提出页岩气井的邻接矩阵构建方法,解决目前考虑多井干扰仅限于直井注采井网的难题,采用图卷积神经网络(GCN)挖掘页岩气井之间的空间特征;其次,基于门控循环单元(GRU)提取不同时间上产量的变化特征,即时间特征,从而形成考虑时空特征的页岩气井机器学习方法;并与数值模拟的结果进行对比验证。最后,基于实际页岩气井的生产井台,对比不同邻接矩阵的构建方法,同时与传统的机器学习方法(LSTM、GRU和RNN)对比。结果表明:与数值模拟结果对比验证了该方法的准确性;在考虑裂缝展布的基础上结合储层渗透率,进一步提高了预测的精度;存在井间干扰的页岩气井采用本模型的精度更高,可超过90%。

    Abstract

    In order to predict shale gas well production more accurately and efficiently, considering 3D well networks and multi-stage hydraulic fracturing horizontal wells in advanced shale gas reservoir development, a novel IMWs-GCN-GRU coupled machine learning method was proposed considering hydraulic fracture distribution and inter-well interference. Firstly, a method for constructing the adjacency matrix was proposed based on a comprehensive consideration of various factors such as the spatial positions of wells, fracture distribution, and reservoir permeability, which can address the limitation of current approaches that considers inter-well interference only within vertical well injection and production networks. A graph convolutional neural network (GCN) technique was employed to explore spatial features among shale gas wells. Secondly, a gate recurrent unit (GRU) method was utilized to extract temporal features over different time periods, thus a novel shale gas well machine learning approach to predict production was formulated that considers both spatial and temporal characteristics. The accuracy of the new model was validated in comparisons with conventional numerical simulation results. Finally, based on the actual production wellhead of shale gas wells, different adjacency matrix construction methods were compared and analyzed. The results show that incorporating fracture distribution along with reservoir permeability on the basis of adjacency matrix construction can further enhance the prediction accuracy of the new model. A comparison with traditional machine learning methods such as LSTM, GRU, and RNN demonstrates that this model can achieve higher accuracy, surpassing 90%, for shale gas wells with inter-well interference.

  • 页岩气作为重要的清洁能源,对保障国家能源安全和“碳达峰、碳中和”目标实现具有重要意义[1]。准确的产量预测是页岩气井全生命周期高效开采的基础。常规的页岩气产量预测方法包括3类:产量递减法方法简单、应用广泛,适用于递减阶段的井[2-4];解析法通过地层参数作出预测,但会结合油藏问题作出理想化假设[5-6];数值模拟法刻画地层精细,能够提供良好的结果,但模拟过程复杂费时[7-8]。近年来,随着人工智能与大数据技术的迅猛发展,基于机器学习方法的产量预测模型受到越来越多的关注[9],早期的机器学习方法包括支持向量机(SVM)[10]、BP神经网络[11]、循环神经网络(RNN)[12]、长短时记忆神经网络(LSTM)[13]、门控循环单元(GRU)[14]等方法;后来,为了提高机器学习方法的速度和全局收敛性,提出了大量耦合模型[15-18],但模型难以考虑井网的空间相关性。近年来,部分学者[19-21]考虑时空特征解决了直井注采井网的产量预测,但针对压裂水平井模型尚未见报道。目前随着页岩气藏的立体井网开采,结合大规模水力压裂,页岩气井之间存在较明显的井间干扰[22-25]。笔者在前人研究的基础上,考虑页岩气井人工裂缝展布和多井干扰特征,提出一种考虑时空特征的IMWs-GCN-GRU耦合页岩气井产量预测方法,利用数值模拟结果对模型进行验证,并采用实际页岩气井井网生产数据,对不同邻接矩阵构建方式与传统时间序列预测方法进行对比。

  • 1 考虑人工裂缝展布和多井干扰的GCN-GRU耦合模型

  • 页岩气井的IMWs-GCN-GRU产量预测耦合模型主要由井间干扰矩阵构建、空间特征提取模块和时间特征提取模块构成。为了更好地描述模型的建立过程,将页岩气井产量预测问题定义为

  • xt+1,xt+2,,xt+T=f(G,X).
    (1)
  • 其中

  • G= (V, E, W) , V=v1, v2, , vN.

  • 式中,[xt+1xt+2,···,xt+T]为时间t+1,t+2,···,t+T的产量预测值;f表征耦合模型网络的变换关系;V为一个井组,每口井假定一个节点viN为井组中单井数量;E为节点之间连接的边;W为对应边E上的权值。

  • 用邻接矩阵A表征各井之间的干扰程度,矩阵A中的值越大则表明两点关系越强,则两口井存在干扰影响越明显;页岩气井历史产量矩阵X

  • X=x11x12x13x1ix21x22x23x2ix31x32x33x3ixt1xt2xt3xti.
    (2)
  • 式中,xti为第i口井在时间t时刻的产量值。

  • 1.1 井间干扰矩阵构建

  • 井与井之间紧密的间距是导致井间干扰发生的根本原因之一,井距越小,干扰现象越明显,同时长水平井结合多段压裂改造形成的水力裂缝,与天然裂缝相交可形成复杂缝网系统,影响井间的连通特性[23];其次,储层的渗透率也会影响井间干扰程度[26-27]

  • 考虑井间干扰时,通常是把一口井作为一个节点[18-21],用于直井的井间干扰分析,但难以表征不同页岩气井长水平段及多段压裂的空间展布关系。本文中提出以压裂水平井的人工裂缝作为节点,建立页岩气井之间的井间干扰程度表征方法。

  • 以两口页岩气井为例,气井1和气井2位于不同的空间位置(图1),气井1有n条人工裂缝,气井2有m条人工裂缝,F11F12F1n为气井1的第1、2、···、n条裂缝;同理,F21F22F2m为气井2的第1、2、···、m条裂缝。

  • 井间干扰程度表征方法如下:以人工裂缝作为节点进行研究,对于存在干扰的两口压裂水平井,其井间干扰程度与各人工缝之间的距离、地层平均渗透率有关,以裂缝F11为例(图1),其受到气井2的干扰程度可以表示为k-L11k-L12k-L13k-L1m

  • 故气井1的n条裂缝之间的干扰程度可表示为k-L11k-L12k-L13k-L1mk-L21k-L22k-L23k-L2mk-L31k-L32k-L33k-L3mk-Ln1k-Ln2k-Ln3k-Lnm

  • 图1 页岩气井之间空间展布示意图

  • Fig.1 Spatial distribution of shale gas wells

  • 将气井各人工裂缝的数值进行相加,得到气井1和气井2之间的井间干扰程度为

  • w1-2=w2-1=i=1n j=1m k-Lij.
    (3)
  • 式中,k-为地层的平均渗透率;Lij为不同井裂缝ij之间的距离;w1-2w2-1为两口井之间的干扰程度,如果井数比较多,则方法类似。若井场有N口井,则多井的邻接矩阵可以表示为

  • A=0w1-2w1-3w1-Nw2-10w2-3w2-Nw3-1w3-20w3-NwN-1wN-2wN-30.
    (4)
  • 1.2 空间特征提取模块

  • 传统的卷积神经网络(CNN)可以获取局部空间特征,但它仅适用于欧氏空间的数据,如图像或规则网格[28]。在实际生活中普遍存在一种非欧空间数据,即图数据,如化学分子、社会网络和交通网络等,该类数据中每个节点的局部特征各异,传统的CNN难以处理。为了解决上述问题,Bruna等[29]提出了图卷积网络(GCN),它可以处理任意图结构数据,由于油气生产的井网开采具有明显的图结构特征,因此本文中采用GCN处理页岩气井的空间结构特征。

  • GCN的核心思想是通过给定的邻接矩阵和特征矩阵聚合周围节点的特征,用于更新自身节点。如图2所示,图卷积网络的基本架构相似,其在各层之间的传播模式可以用数学方式表示为

  • H(l+1)=σD^-12A^D^-12H(l)W(l).
    (5)
  • 式中,A^=A+I为一个自连接的邻接矩阵;I为单位矩阵;D^=j A^ij为度矩阵;HlHl+1)分别为第ll+1层的特征输出;Wl为第l层权重矩阵;σ为非线性激活函数。

  • 图2 GCN网络基本框架

  • Fig.2 GCN network structure

  • 1.3 时间特征提取模块

  • 门控循环单元(GRU)是长短时记忆神经网络(LSTM)的变体,Cho等[30-31]为了解决梯度消失问题同时保留时间序列特征将LSTM的遗忘门和输入门整合为更新门,提高了训练速度。GRU的结构单元如图3所示,其向

  • rt=σWrht-1,xt+br,zt=σWzht-1,xt+bz,h~t=tanhWhht-1rt,xt+bh,ht=1-ztht-1+zth~t.
    (6)
  • 图3 GRU结构单元

  • Fig.3 GRU structural unit

  • 式中,rtzt分别为重置门和更新门;xtht-1h~tht分别为t时的输入、t-1时的输出、更新后的单元状态和t时的输出;WrWzWh分别为重置门、更新门和当前记忆内容的权重矩阵;brbzbh分别为重置门、更新门和当前记忆内容的偏置矩阵;⊙为Hadamard积(按元素乘积)运算符;σ、tanh分别表示sigmoid和tanh激活函数。

  • 1.4 页岩气井GCN-GRU耦合产量预测模型

  • 在考虑页岩气压裂水平井之间人工裂缝展布特征和井间干扰的基础上,分别采用GCN网络处理井间复杂的空间相关性,同时GRU捕捉历史产量数据的时间特征,提出一种新的IMWs-GCN-GRU耦合页岩气井产量预测模型。

  • 模型结构如图4所示,该模型主要由井间干扰矩阵构建模块、空间特征提取模块和时间特征提取模块构成。首先,通过页岩气压裂水平井裂缝的空间分布关系和储层的地质参数来定义井间干扰程度,由此构建出表征井间干扰的邻接矩阵A;其次将井间干扰邻接矩阵A和历史产量数据作为该模型的输入,使用GCN提取产量数据中的空间特征;再将得到的具有空间特征的序列输入到GRU中以提取时间特征;最后使用全连接层得到页岩气井产量预测结果。

  • 图4 IMWs-GCN-GRU耦合模型

  • Fig.4 Structure of IMWs-GCN-GRU coupling model

  • 1.5 评价指标

  • 为了对比模型的准确性,采用均方根误差(root mean square error,RMSE)RRMSE、平均绝对误差(mean absolute error,MAE)RMAE、预测准确率(forecast accuracy,ACC)RACC和平均相对误差(mean relative error,MRE)RMRE 4种评价指标评价模型的预测效果。模型的RRMSERMAERMRE的值越小,且RACC的值越接近1,说明模型的预测效果越好,

  • RRMSE=1ni=1n yi-y^i2,
    (7)
  • RMAE=1ni=1n yi-y^i,
    (8)
  • RACC=1-1ni=1n yi-y^iyi,
    (9)
  • RMRE=1ni=1n yi-y^iyi.
    (10)
  • 式中,yi为原始产量数据; y^i为预测值; n 为预测的样本数量。

  • 2 模型验证及因素分析

  • 为了验证提出的IMWs-GCN-GRU耦合模型的可行性及准确性,使用数值模拟方法构建页岩气压裂水平井理论模型,分别从气井的不同平面和纵向展布进行对比分析,设计平行分布、交错分布和完全错开分布3种不同的布井方式。

  • 采用数值模拟方法模拟两口页岩气井生产过程,气井1和气井2的水平段长1000 m,裂缝5条,裂缝开度为0.003 m,裂缝半长分别为25和50 m,裂缝间距为200 m。先定产量生产,产量设置为25000 m3/d;后定生产压差生产,压力设置为5 MPa。生产时间为5 a,页岩气井日产气量为产量测试及预测数据。

  • 2.1 不同的平面展布

  • 针对气井不同平面展布的情况,分别讨论平行分布、交错分布和完全错开分布3种不同的布井方案,如图5和6所示。两口井水平段分布在同一小层,该储层渗透率为5.00×10-5 μm2。水平段沿同一方向延伸,两口井的井距设置为200 m。

  • 图5 气井不同的平面展布示意图

  • Fig.5 Schematic diagram of different plane layouts of gas wells

  • 图6 不同平面布井方式产量对比

  • Fig.6 Production comparison of different plane well layout methods

  • 由图6可见,随着错开程度的增加单井控制储量增加,气井产量有明显的提升,且错开程度越大,气井产量递减的速率越慢。依据两口井的位置分布关系、裂缝参数以及地层渗透率,采用井间干扰矩阵构建方法得到的邻接矩阵为

  • A=00.7540.7540, ; 00.6710.6710, ; 00.3420.3420, .
    (11)
  • 由式(11)可见两井之间邻接关系越密切,邻接矩阵值越大,可见邻接矩阵值在一定程度上代表井间的干扰程度。将邻接矩阵和两口井的产量数据按照训练集和测试集数量8∶2的方式,作为IMWs-GCN-GRU耦合模型的输入,得到气井产量的预测结果如表1所示。3种不同的布井方案下,气井产量误差非常小,平均相对误差均小于0.1%,故该模型在气井平面上不同布井方案时产量预测结果准确可行。

  • 表1 平面上不同展布两口井产量预测误差

  • Table1 Production prediction errors of different planar distribution

  • 2.2 不同的立体展布

  • 为了进一步说明模型的准确性及适应性,在平面上不同布井方案的基础上构建一个纵向上从1~10随机生成的非均质地层,从1到10层的渗透率随机分布为[2.4,6.05,1.94,5.77,8.67,7.94,1.3,8.19,4.12,5.8],把两口井布置在不同的层位进行研究,如图7所示。与2.1节中类似,两口井的井距设置为200 m,纵向上距离为20 m。气井1分布在第4层,平均渗透率为5.77×10-5μm2,气井2在第6层平均渗透率为7.94×10-5μm2

  • 通过数值模拟两口气井的产量数据如图8所示。与平面展布相比,由于渗透率的非均质变化,气井的产量发生了一定的变化,特别是气井2,由于渗透率的增加,稳产时间更长,产量递减明显减缓。

  • 此时,两口井的邻接矩阵为

  • A=00.5610.5610, ; 00.5010.5010, ; 00.2560.2560, .
    (12)
  • 图7 气井不同的立体展布

  • Fig.7 Different 3D distribution of gas wells

  • 图8 不同立体展布产量计算结果

  • Fig.8 Production of different 3D distribution of gas wells

  • 由式(12)可见,相对于平面展布情况,立体展布时两井的干扰程度减弱,邻接矩阵的值也相应地减小。同理采用IMWs-GCN-GRU耦合模型对两口页岩气井的产量进行预测,误差较小,相对误差小于0.2%(表2),说明该方法对非均质地层仍然适用。

  • 表2 不同立体展布两口井产量预测误差

  • Table2 Production prediction errors of different 3D distribution

  • 3 实例分析

  • 选取一个实际生产的页岩气压裂水平井井网进行研究。首先,对比3种不同的邻接矩阵构建方案,以此说明本模型邻接矩阵构建方法的准确性;其次,将该模型与常规时间序列模型进行对比,包括LSTM、GRU和RNN,检验IMWs-GCN-GRU耦合模型的预测精度。

  • 3.1 数据收集及预处理

  • 实际页岩气井台包含6口生产井,井位如图9所示。其中气井1、气井3和气井4相距较近,气井2和气井5相距较近,同时气井6离其他5口井较远。收集各页岩气井从2020年6月投产至2022年5月的产量数据,如图10所示。由图10可见,在生产过程中每口井的生产制度都发生了一定的改变,产量曲线波动明显,增加了预测的难度。在进行产量预测之前需要对数据集进行预处理,将原始产量数据的前80%划分为训练集,将后20%划分为测试集,并分别对训练集和测试集进行归一化处理。

  • 图9 气井井位及裂缝分布

  • Fig.9 Fracture distribution and position of shale gas wells

  • 图10 气井的原始产量数据

  • Fig.10 Raw production data of shale gas wells

  • 3.2 模型预测

  • 为了实现多井干扰时的页岩气井产量预测,使用Python3.7基于Tensorflow1.15.0对模型进行编译。同时,对模型的超参数进行优化,在50、100、150、200、250、300之间选择门控循环单元个数,并分析预测精度的变化,如图11所示。门控循环单元个数为150时,模型的预测效果最佳,确定最终模型的参数设置:学习率为0.001,输入步长为2,预测步长为1,迭代次数为2000,批处理大小为128,门控循环单元个数为150。

  • 图11 门控循环单元个数对预测精度的影响

  • Fig.11 Impact of number of gate recurrent units on prediction accuracy

  • 3.3 不同邻接矩阵构建方式结果对比

  • 3.3.1 不同的邻接矩阵构建方法

  • 为了说明所构建的邻接矩阵高效性,采用3种方案对比不同邻接矩阵构建方法下页岩气井的产量预测结果。

  • (1)方案一为最常用的邻接矩阵构建方法,即两点若相关,邻接矩阵值为1;若不相关,则为0[32]

  • (2)方案二是依据井距构建邻接矩阵,则表征第k口井受第l口井干扰程度的邻接矩阵的值为

  • wkl=wlk=i=1n j=1n 1Lij.
    (13)
  • 此时页岩气井井网的邻接矩阵为

  • A=00.1770.3080.4250.1930.0730.17700.2100.2590.7480.1340.3080.21000.8560.2670.1000.4250.2590.85600.3470.1090.1930.7480.2670.34700.1340.0730.1340.1000.1090.1340.
    (14)
  • (3)方案三是考虑井距和地层平均渗透率来构建邻接矩阵,邻接矩阵的值为

  • wkl=wlk=i=1n j=1n k-Lij.
    (15)
  • 得到的实际井网邻接矩阵为

  • A=00.1860.3430.4720.1980.0740.18600.2340.2880.7640.1370.3430.23401.0000.2880.1080.4720.2881.00000.3740.1170.1980.7640.2880.37400.1330.0740.1370.1080.1170.1330.
    (16)
  • 3.3.2 预测结果

  • 分别使用以上3种方案所构建的邻接矩阵对6口页岩气井进行产量预测,并计算各井的评价指标。表3是3种方案下6口井的平均预测误差,对比误差可以发现,方案一相较于其他两种方案预测效果最差,其RMSE和MAE值最高,ACC值最低。因此在构建邻接矩阵时考虑井距和渗透率,可以提高预测的精度。进一步对比方案二、三可见,方案三的评价指标最好,预测效果优于其他方案。

  • 表3 三种方案预测误差

  • Table3 Prediction errors of three schemes

  • 图12为3种方案下6口井的产量预测部分结果,通过对比可见,方案三在6口井中的预测结果更接近实际产量。其次,气井3在红色区域(预测第82 d)生产措施发生改变,产量曲线出现明显下降,由于井间干扰的影响,方案一、二和三中不同井的预测结果在80 d左右都出现了不同程度的产量下降,说明考虑邻接矩阵的IMWs-GCN-GRU模型可以反映页岩气井之间的干扰现象。方案一在构建邻接矩阵时,并未考虑井间的干扰程度,因此其6口井的预测结果在红色区域均出现了产量的下降,与实际数据不符。而方案二在构建邻接矩阵时考虑井距,一定程度上表征了井间的干扰程度,方案二中的气井2、气井5和气井6的预测结果在82 d左右并未出现突然下降。方案三在方案二的基础上,进一步考虑了地层渗透率的影响,气井1、气井2和气井5的预测结果相较于方案二更加符合实际数据。

  • 图12 不同邻接矩阵构成方案的预测结果

  • Fig.12 Predicted results for different adjacency matrix-based schemes

  • 3.4 不同网络模型产量预测结果对比

  • 将本文中模型与常见的时间序列模型进行对比,包括LSTM、GRU和RNN,不同网络模型在气井3的预测结果如图13所示。各井的预测误差结果见表4。由图13可见,IMWs-GCN-GRU模型的页岩气井产量预测结果更加准确,LSTM模型可得到气井3预测82 d左右产量降低,但后期结果无法稳定;GRU和RNN预测结果为平滑的曲线,没能捕捉到产量的降低过程。

  • 由表4可见:IMWs-GCN-GRU模型在气井1至气井5中的预测效果最好,其评价指标是所有模型中最优的,RMSE值各井平均降低了41.95%,RMAE值各井平均降低了40.96%,ACC值各井平均提高了10.15%,由于气井1至气井5之间存在一定的井间干扰,在IMWs-GCN-GRU模型的输入中有表征井间干扰的邻接矩阵,预测时能够考虑到干扰井所带来的影响,而LSTM、GRU和RNN的输入为原始的产量数据,无法考虑井间干扰导致的产量变化,因此当页岩气井间存在井间干扰时,IMWs-GCN-GRU模型可以有效提高预测精度。而在气井6中,预测结果最好的模型为GRU模型,其次是LSTM模型和RNN模型,最差为IMWs-GCN-GRU模型,主要是由于气井6与其他井之间干扰十分微弱,邻接矩阵的图卷积构建反而对产量的预测进行了干扰,因此页岩气井井距较大时可以忽略井间干扰的影响,常见的时间序列模型预测效果更好。

  • 表4 各网络模型在各井的预测误差

  • Table4 Prediction errors of various network models for each well

  • 图13 不同网络模型在气井3的预测结果

  • Fig.13 Prediction results of different network models for gas well 3

  • 4 结论

  • (1)基于提出的井间干扰邻接矩阵构建模块,通过各人工缝之间的距离、地层平均渗透率等参数表征井间干扰程度,井间干扰程度越大,邻接矩阵值越大。

  • (2)与页岩气井数值模拟结果对比,平面上不同布井方式和纵向非均质不同布井方式下的页岩气井产量预测结果,本模型的预测精度高,相对误差小于1%,模型具备准确性和可行性。

  • (3)邻接矩阵中引入井间空间位置分布、裂缝参数、地层平均渗透率等参数有助于提高预测精度。

  • (4)对于存在相互干扰的井,IMWs-GCN-GRU耦合模型对气井产量的预测结果明显优于常规时间序列机器学习模型(LSTM、GRU、RNN),但对于井距较远未发生干扰的井,常规时间序列机器学习模型更优。

  • 参考文献

    • [1] 张君峰,周志,宋腾,等.中美页岩气勘探开发历程、地质特征和开发利用条件对比及启示[J].石油学报,2022,43(12):1687-1701.ZHANG Junfeng,ZHOU Zhi,SONG Teng,et al.Comparison of exploration and development history,geological characteristics and exploitation conditions of shale gas in China and the United States and its enlightenment[J].Acta Petrolei Sinica,2022,43(12):1687-1701.

    • [2] 赵康,陈民锋,王艺文,等.四川盆地SZ页岩气藏气井产量变化规律及递减预测新模型[J].中国海上油气,2024,36(3):129-136.ZHAO Kang,CHEN Minfeng,WANG Yiwen,et al.A new model for production variation patterns and decline prediction of SZ shale gas reservoir in Sichuan Basin[J].China Offshore Oil and Gas,2024,36(3):129-136.

    • [3] 王睿,薛龙龙,党东琪,等.产量递减方程通式的建立及应用[J].石油学报,2023,44(10):1693-1705,1738.WANG Rui,XUE Longlong,DANG Dongqi,et al.Establishment and application of general equation for production decline[J].Acta Petrolei Sinica,2023,44(10):1693-1705,1738.

    • [4] ARPS J J.Analysis of decline curves[J].Transactions of the AIME,2016,160:228-247.

    • [5] 徐兵祥,李相方,HAGHIGHI Manouchehr,等.裂缝性页岩气藏水平井产能预测模型[J].开云电竞投注学报(自然科学版),2013,37(6):92-99,105.XU Bingxiang,LI Xiangfang,HAGHIGHI Manouchehr,et al.A new model for production analysis in naturally fractured shale gas reservoirs[J].Journal of China University of Petroleum(Edition of Natural Science),2013,37(6):92-99,105.

    • [6] 陈元千,王鑫,刘洋.Blasingame气井变产量拟稳态压降方程存在的问题及其推导与应用[J].中国海上油气,2023,35(6):70-77.CHEN Yuanqian,WANG Xin,LIU Yang.Existing problems of Blasingames pseudo-steady-state pressure drawdown equation with variable rate for gas wells along with derivation and application[J].China Offshore Oil and Gas,2023,35(6):70-77.

    • [7] 张黎明,陈昕晟,李国欣,等.基于集合和神经网络架构搜索的自动历史拟合方法[J].开云电竞投注学报(自然科学版),2022,46(2):127-136.ZHANG Liming,CHEN Xinsheng,LI Guoxin,et al.An automatic history matching method based on ensemble and neural architecture search[J].Journal of China University of Petroleum(Edition of Natural Science),2022,46(2):127-136.

    • [8] CLARKSON C R R,NOBAKHT M,KAVIANI D,et al.Production analysis of tight-gas and shale-gas reservoirs using the dynamic-slippage concept[J].SPE Journal,2012,17(1):230-242.

    • [9] 林伯韬,郭建成.人工智能在石油工业中的应用现状探讨[J].石油科学通报,2019,4(4):403-413.LIN Botao,GUO Jiancheng.Discussion on current application of artificial intelligence in petroleum industry[J].Petroleum Science Bulletin,2019,4(4):403-413.

    • [10] 杨磊,王化增,陈子凌.基于支持向量机的油气储量价值等级评价[J].开云电竞投注学报(自然科学版),2012,36(3):192-196.YANG Lei,WANG Huazeng,CHEN Ziling.Assessing value classification of oil and gas reserve based on support vector machine[J].Journal of China University of Petroleum(Edition of Natural Science),2012,36(3):192-196.

    • [11] 周彩兰,刘敏.BP神经网络在石油产量预测中的应用[J].武汉理工大学学报,2009,31(3):125-129.ZHOU Cailan,LIU Min.Application research on oil production forecasting based on BP neural network[J].Journal of Wuhan University of Technology,2009,31(3):125-129.

    • [12] 高亚军,唐力辉,王振鹏,等.基于循环神经网络和数据差分处理的油田产量预测方法[J].中国海上油气,2023,35(3):126-136.GAO Yajun,TANG Lihui,WANG Zhenpeng,et al.A method for oilfield production prediction based on recurrent neural network and data differential processing[J].China Offshore Oil and Gas,2023,35(3):126-136.

    • [13] 刘合,李艳春,杜庆龙,等.基于多变量时间序列模型的高含水期产量预测方法[J].开云电竞投注学报(自然科学版),2023,47(5):103-114.LIU He,LI Yanchun,DU Qinglong,et al.Prediction of production during high water-cut period based on multivariate time series model[J].Journal of China University of Petroleum(Edition of Natural Science),2023,47(5):103-114.

    • [14] LI X,MA X,XIAO F,et al.Application of gated recurrent unit(GRU)neural network for smart batch production prediction[J].Energies,2020,13(22):6121.

    • [15] 樊冬艳,杨灿,孙海,等.基于时间序列相似性与机器学习方法的页岩气井产量预测[J].开云电竞投注学报(自然科学版),2024,48(3):119-126.FAN Dongyan,YANG Can,SUN Hai,et al.Shale gas well production forecasting based on time sequence similarity and machine learning methods[J].Journal of China University of Petroleum(Edition of Natural Science),2024,48(3):119-126.

    • [16] ZHA W,LIU Y,WAN Y,et al.Forecasting monthly gas field production based on the CNN-LSTM model[J].Energy,2022,260:124889.

    • [17] 姜宝胜,白玉湖,徐兵祥,等.基于集成学习的致密气藏产能预测新方法[J].中国海上油气,2024,36(5):120-127.JIANG Baosheng,BAI Yuhu,XU Bingxiang,et al.A novel approach for predicting production capacity of tight gas reservoirs based on ensemble learning[J].China Offshore Oil and Gas,2024,36(5):120-127.

    • [18] FAN D,SUN H,YAO J,et al.Well production forecasting based on ARIMA-LSTM model considering manual operations[J].Energy,2021,220:119708.

    • [19] GAO M,WEI C,ZHAO X,et al.Production forecasting based on attribute-augmented spatiotemporal graph convolutional network for a typical carbonate reservoir in the middle east[J].Energies,2023,16(1):407.

    • [20] DU E,LIU Y,CHENG Z,et al.Production forecasting with the interwell interference by integrating graph convolutional and long short-term memory neural network[J].SPE Reservoir Evaluation & Engineering,2022,25(2):197-213.

    • [21] DU E,LIU Y,XUE L,et al.A data-driven model for production prediction of strongly heterogeneous reservoir under uncertainty[J].Geoenergy Science and Engineering,2023,223:211542.

    • [22] 郭旭洋,金衍,黄雷,等.页岩油气藏水平井井间干扰研究现状和讨论[J].石油钻采工艺,2021,43(3):348-367.GUO Xuyang,JIN Yan,HUANG Lei,et al.Research status and discussion of horizontal well interference in shale oil and gas reservoirs[J].Oil Drilling & Production Technology,2021,43(3):348-367.

    • [23] AL-SHAMI T M,JUFAR S R,KUMAR S,et al.A comprehensive review of interwell interference in shale reservoirs[J].Earth-Science Reviews,2023,237:104327.

    • [24] 程时清,李猛,何佑伟,等.一种用于多级压裂水平井判别来水方向的多井干扰压力分析方法[J].开云电竞投注学报(自然科学版),2018,42(5):81-88.CHENG Shiqing,LI Meng,HE Youwei,et al.A multi-well interference pressure transient analysis method to determine the water source orientation of multi-fractured horizontal well[J].Journal of China University of Petroleum(Edition of Natural Science),2018,42(5):81-88.

    • [25] PATEL H,CADWALLADER S,WAMPLER J.Zipper fracturing:taking theory to reality in the eagle ford shale[R].SPE 2445923-MS 2016.

    • [26] 鹿克峰,苏畅,程超逸.一种评价封闭气藏采气指数、动储量及井间连通性的简便方法[J].中国海上油气,2023,35(4):86-93.LU Kefeng,SU Chang,CHENG Chaoyi.A simple method for evaluating gas production index,dynamic reserves and interwell connectivity of closed gas reservoirs[J].China Offshore Oil and Gas,2023,35(4):86-93.

    • [27] LAWAL H,ABOLO N,JACKSON G,et al.A quantitative approach to analyze fracture area loss in shale gas reservoirs[R].SPE 169406-MS,2014.

    • [28] ZHAO L,SONG Y,ZHANG C,et al.T-GCN:a temporal graph convolutional network for traffic prediction[J].IEEE Transactions on Intelligent Transportation Systems,2019,21(9):3848-3858.

    • [29] BRUNA J,ZAREMBA W,SZLAM A,et al.Spectral networks and locally connected networks on graphs[J].Computer Science,2013,arXiv:1312.6203.

    • [30] CHO K,VAN MERRIENBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation [J].Computer Science,2014,arXiv:1406.1078

    • [31] 杨丽,吴雨茜,王俊丽,等.循环神经网络研究综述[J].计算机应用,2018,38(增刊2):1-6,26.YANG Li,WU Yuqian,WANG Junli,et al.Research on recurrent neural network[J].Journal of Computer Applications,2018,38(sup2):1-6,26.

    • [32] HARARY F.The determinant of the adjacency matrix of a graph[J].SIAM Review,1962,4(3):202-210.

  • 参考文献

    • [1] 张君峰,周志,宋腾,等.中美页岩气勘探开发历程、地质特征和开发利用条件对比及启示[J].石油学报,2022,43(12):1687-1701.ZHANG Junfeng,ZHOU Zhi,SONG Teng,et al.Comparison of exploration and development history,geological characteristics and exploitation conditions of shale gas in China and the United States and its enlightenment[J].Acta Petrolei Sinica,2022,43(12):1687-1701.

    • [2] 赵康,陈民锋,王艺文,等.四川盆地SZ页岩气藏气井产量变化规律及递减预测新模型[J].中国海上油气,2024,36(3):129-136.ZHAO Kang,CHEN Minfeng,WANG Yiwen,et al.A new model for production variation patterns and decline prediction of SZ shale gas reservoir in Sichuan Basin[J].China Offshore Oil and Gas,2024,36(3):129-136.

    • [3] 王睿,薛龙龙,党东琪,等.产量递减方程通式的建立及应用[J].石油学报,2023,44(10):1693-1705,1738.WANG Rui,XUE Longlong,DANG Dongqi,et al.Establishment and application of general equation for production decline[J].Acta Petrolei Sinica,2023,44(10):1693-1705,1738.

    • [4] ARPS J J.Analysis of decline curves[J].Transactions of the AIME,2016,160:228-247.

    • [5] 徐兵祥,李相方,HAGHIGHI Manouchehr,等.裂缝性页岩气藏水平井产能预测模型[J].开云电竞投注学报(自然科学版),2013,37(6):92-99,105.XU Bingxiang,LI Xiangfang,HAGHIGHI Manouchehr,et al.A new model for production analysis in naturally fractured shale gas reservoirs[J].Journal of China University of Petroleum(Edition of Natural Science),2013,37(6):92-99,105.

    • [6] 陈元千,王鑫,刘洋.Blasingame气井变产量拟稳态压降方程存在的问题及其推导与应用[J].中国海上油气,2023,35(6):70-77.CHEN Yuanqian,WANG Xin,LIU Yang.Existing problems of Blasingames pseudo-steady-state pressure drawdown equation with variable rate for gas wells along with derivation and application[J].China Offshore Oil and Gas,2023,35(6):70-77.

    • [7] 张黎明,陈昕晟,李国欣,等.基于集合和神经网络架构搜索的自动历史拟合方法[J].开云电竞投注学报(自然科学版),2022,46(2):127-136.ZHANG Liming,CHEN Xinsheng,LI Guoxin,et al.An automatic history matching method based on ensemble and neural architecture search[J].Journal of China University of Petroleum(Edition of Natural Science),2022,46(2):127-136.

    • [8] CLARKSON C R R,NOBAKHT M,KAVIANI D,et al.Production analysis of tight-gas and shale-gas reservoirs using the dynamic-slippage concept[J].SPE Journal,2012,17(1):230-242.

    • [9] 林伯韬,郭建成.人工智能在石油工业中的应用现状探讨[J].石油科学通报,2019,4(4):403-413.LIN Botao,GUO Jiancheng.Discussion on current application of artificial intelligence in petroleum industry[J].Petroleum Science Bulletin,2019,4(4):403-413.

    • [10] 杨磊,王化增,陈子凌.基于支持向量机的油气储量价值等级评价[J].开云电竞投注学报(自然科学版),2012,36(3):192-196.YANG Lei,WANG Huazeng,CHEN Ziling.Assessing value classification of oil and gas reserve based on support vector machine[J].Journal of China University of Petroleum(Edition of Natural Science),2012,36(3):192-196.

    • [11] 周彩兰,刘敏.BP神经网络在石油产量预测中的应用[J].武汉理工大学学报,2009,31(3):125-129.ZHOU Cailan,LIU Min.Application research on oil production forecasting based on BP neural network[J].Journal of Wuhan University of Technology,2009,31(3):125-129.

    • [12] 高亚军,唐力辉,王振鹏,等.基于循环神经网络和数据差分处理的油田产量预测方法[J].中国海上油气,2023,35(3):126-136.GAO Yajun,TANG Lihui,WANG Zhenpeng,et al.A method for oilfield production prediction based on recurrent neural network and data differential processing[J].China Offshore Oil and Gas,2023,35(3):126-136.

    • [13] 刘合,李艳春,杜庆龙,等.基于多变量时间序列模型的高含水期产量预测方法[J].开云电竞投注学报(自然科学版),2023,47(5):103-114.LIU He,LI Yanchun,DU Qinglong,et al.Prediction of production during high water-cut period based on multivariate time series model[J].Journal of China University of Petroleum(Edition of Natural Science),2023,47(5):103-114.

    • [14] LI X,MA X,XIAO F,et al.Application of gated recurrent unit(GRU)neural network for smart batch production prediction[J].Energies,2020,13(22):6121.

    • [15] 樊冬艳,杨灿,孙海,等.基于时间序列相似性与机器学习方法的页岩气井产量预测[J].开云电竞投注学报(自然科学版),2024,48(3):119-126.FAN Dongyan,YANG Can,SUN Hai,et al.Shale gas well production forecasting based on time sequence similarity and machine learning methods[J].Journal of China University of Petroleum(Edition of Natural Science),2024,48(3):119-126.

    • [16] ZHA W,LIU Y,WAN Y,et al.Forecasting monthly gas field production based on the CNN-LSTM model[J].Energy,2022,260:124889.

    • [17] 姜宝胜,白玉湖,徐兵祥,等.基于集成学习的致密气藏产能预测新方法[J].中国海上油气,2024,36(5):120-127.JIANG Baosheng,BAI Yuhu,XU Bingxiang,et al.A novel approach for predicting production capacity of tight gas reservoirs based on ensemble learning[J].China Offshore Oil and Gas,2024,36(5):120-127.

    • [18] FAN D,SUN H,YAO J,et al.Well production forecasting based on ARIMA-LSTM model considering manual operations[J].Energy,2021,220:119708.

    • [19] GAO M,WEI C,ZHAO X,et al.Production forecasting based on attribute-augmented spatiotemporal graph convolutional network for a typical carbonate reservoir in the middle east[J].Energies,2023,16(1):407.

    • [20] DU E,LIU Y,CHENG Z,et al.Production forecasting with the interwell interference by integrating graph convolutional and long short-term memory neural network[J].SPE Reservoir Evaluation & Engineering,2022,25(2):197-213.

    • [21] DU E,LIU Y,XUE L,et al.A data-driven model for production prediction of strongly heterogeneous reservoir under uncertainty[J].Geoenergy Science and Engineering,2023,223:211542.

    • [22] 郭旭洋,金衍,黄雷,等.页岩油气藏水平井井间干扰研究现状和讨论[J].石油钻采工艺,2021,43(3):348-367.GUO Xuyang,JIN Yan,HUANG Lei,et al.Research status and discussion of horizontal well interference in shale oil and gas reservoirs[J].Oil Drilling & Production Technology,2021,43(3):348-367.

    • [23] AL-SHAMI T M,JUFAR S R,KUMAR S,et al.A comprehensive review of interwell interference in shale reservoirs[J].Earth-Science Reviews,2023,237:104327.

    • [24] 程时清,李猛,何佑伟,等.一种用于多级压裂水平井判别来水方向的多井干扰压力分析方法[J].开云电竞投注学报(自然科学版),2018,42(5):81-88.CHENG Shiqing,LI Meng,HE Youwei,et al.A multi-well interference pressure transient analysis method to determine the water source orientation of multi-fractured horizontal well[J].Journal of China University of Petroleum(Edition of Natural Science),2018,42(5):81-88.

    • [25] PATEL H,CADWALLADER S,WAMPLER J.Zipper fracturing:taking theory to reality in the eagle ford shale[R].SPE 2445923-MS 2016.

    • [26] 鹿克峰,苏畅,程超逸.一种评价封闭气藏采气指数、动储量及井间连通性的简便方法[J].中国海上油气,2023,35(4):86-93.LU Kefeng,SU Chang,CHENG Chaoyi.A simple method for evaluating gas production index,dynamic reserves and interwell connectivity of closed gas reservoirs[J].China Offshore Oil and Gas,2023,35(4):86-93.

    • [27] LAWAL H,ABOLO N,JACKSON G,et al.A quantitative approach to analyze fracture area loss in shale gas reservoirs[R].SPE 169406-MS,2014.

    • [28] ZHAO L,SONG Y,ZHANG C,et al.T-GCN:a temporal graph convolutional network for traffic prediction[J].IEEE Transactions on Intelligent Transportation Systems,2019,21(9):3848-3858.

    • [29] BRUNA J,ZAREMBA W,SZLAM A,et al.Spectral networks and locally connected networks on graphs[J].Computer Science,2013,arXiv:1312.6203.

    • [30] CHO K,VAN MERRIENBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation [J].Computer Science,2014,arXiv:1406.1078

    • [31] 杨丽,吴雨茜,王俊丽,等.循环神经网络研究综述[J].计算机应用,2018,38(增刊2):1-6,26.YANG Li,WU Yuqian,WANG Junli,et al.Research on recurrent neural network[J].Journal of Computer Applications,2018,38(sup2):1-6,26.

    • [32] HARARY F.The determinant of the adjacency matrix of a graph[J].SIAM Review,1962,4(3):202-210.

  • Baidu
    map