用于两相流环空压力预测的自适应物理信息神经网络模型

doi:10.7623/syxb202303012

石油学报 ›› 2023, Vol. 44 ›› Issue (3): 545-555.DOI: 10.7623/syxb202303012

用于两相流环空压力预测的自适应物理信息神经网络模型

徐宝昌¹, 张学智¹, 王雅欣¹, 刘伟², 孟卓然¹

1. 中国石油大学(北京)信息科学与工程学院北京 102249;
2. 中国石油集团工程技术研究院有限公司北京 102206

收稿日期:2021-12-29 修回日期:2023-02-06 出版日期:2023-03-25 发布日期:2023-04-06
通讯作者: 徐宝昌,男,1974年8月生,2005年获北京航空航天大学博士学位,现为中国石油大学(北京)信息科学与工程学院副教授,主要从事钻井工程自动化、复杂系统建模与先进控制、多传感器信息融合与软测量技术方面的研究工作。Email:xbcyl@cup.edu.cn
作者简介:徐宝昌,男,1974年8月生,2005年获北京航空航天大学博士学位,现为中国石油大学(北京)信息科学与工程学院副教授,主要从事钻井工程自动化、复杂系统建模与先进控制、多传感器信息融合与软测量技术方面的研究工作。Email:xbcyl@cup.edu.cn
基金资助:
国家重点研发计划项目(2019YFA0708304)和中国石油天然气集团有限公司-中国石油大学(北京)战略合作科技专项(ZLZX2020-03)资助。

Self-adaptive physical information neural network model for prediction of two-phase flow annulus pressure

Xu Baochang¹, Zhang Xuezhi¹, Wang Yaxin¹, Liu Wei², Meng Zhuoran¹

1. College of Information Science and Engineering, China University of Petroleum, Beijing 102249, China;
2. CNPC Engineering Technology R & D Company Limited, Beijing 102206, China

Received:2021-12-29 Revised:2023-02-06 Online:2023-03-25 Published:2023-04-06

摘要/Abstract

摘要： 石油勘探开发不断面向着更为复杂多变的地层,为应对深层复杂油气藏钻探过程中存在的气侵现象,需要采用控压钻井技术(MPD)以防止气侵导致的井喷事故。其中,通过求解气侵工况下两相流井筒水力学偏微分方程组(PDEs)来准确预测井底压力是制定控制方案的关键。采用基于自适应物理信息神经网络(PINN)方法对两相流井筒环空压力进行预测:首先,根据井筒机理设计全连接神经网络,用于拟合训练数据样本的分布;其次,将已知的两相流井筒模型的偏微分方程以微分形式约束条件融入神经网络的损失函数中,此外,采用可训练的自适应权重提升神经网络模型精度,使网络在训练过程中着重关注边界点、初始点等求解困难区域;最后采用Adam算法对网络参数和微分方程的权重进行优化,并使用L-BFGS算法对网络参数进一步优化。随机选取有限差分法在稠密网格情况下求解井筒水力学模型所得的部分数据作为实验数据集。实验结果表明,相较于常规的物理信息神经网络和传统的有限差分法,用于两相流环空压力预测的自适应物理信息神经网络模型性能更佳。

关键词: 物理信息神经网络, 偏微分方程, 井筒水力学, 压力预测, 自适应

Abstract: With the deepening of oil exploration and development in more and more complex and variable formations, Managed Pressure Drilling (MPD)has to be used for preventing blowouts caused by gas intrusion during drilling in deep and complex reservoirs. Specifically, accurate prediction of bottomhole pressure by solving partial differential equations (PDEs)for the two-phase flowing wellbore under gas intrusion conditions is the key to develop a control scheme. A method based on an adaptive Physics-Informed Neural Networks (PINN)is used to predict the two-phase flow wellbore annulus pressure. First, a fully connected neural network is designed based on the wellbore mechanism, so as to fit the distribution of training data samples; second, the known partial differential equations of the two-phase flow wellbore model are incorporated into the loss function of the neural network as differential constraints. In addition, trainable adaptive weights are used to enhance the accuracy of the neural network model, so that the network can put the focus on boundary points, initial points, and other difficult regions of solution during the training process; finally, the Adam algorithm is used to optimize the network parameters and the weights of differential equations, and the L-BFGS algorithm is used to further optimize the network parameters. Some data obtained from solving the wellbore hydraulics model by finite difference method in the dense grid case are randomly selected as the experimental data set. The experimental results show that the self-adaptive physical information neural network model proposed for prediction of two-phase flow annulus pressure has better performance than the conventional physical information neural network and the traditional finite difference method.

Key words: physical information neural network, partial differential equation, wellbore hydraulics, pressure prediction, self-adaptive

中图分类号:

TE928

徐宝昌, 张学智, 王雅欣, 刘伟, 孟卓然. 用于两相流环空压力预测的自适应物理信息神经网络模型[J]. 石油学报, 2023, 44(3): 545-555.

Xu Baochang, Zhang Xuezhi, Wang Yaxin, Liu Wei, Meng Zhuoran. Self-adaptive physical information neural network model for prediction of two-phase flow annulus pressure[J]. Acta Petrolei Sinica, 2023, 44(3): 545-555.

参考文献

[1] 徐朝阳,孟英峰,魏纳,等.气侵过程井筒压力演变规律实验和模型[J].石油学报,2015,36(1):120-126.
XU Chaoyang,MENG Yingfeng,WEI Na,et al.Experimental simulation and numerical modeling of dynamic variations in wellbore pressure during gas-kicks[J].Acta Petrolei Sinica,2015,36(1):120-126.
[2] 刘伟,周英操,石希天,等.塔里木油田库车山前超高压盐水层精细控压钻井技术[J].石油钻探技术,2020,48(2):23-28.
LIU Wei,ZHOU Yingcao,SHI Xitian,et al.Precise managed pressure drilling technology for ultra-high pressure brine layer in the Kuqa piedmont of the Tarim oilfield[J].Petroleum Drilling Techniques,2020,48(2):23-28.
[3] 徐宝昌,孟宇,刘伟.控压钻井井下不可测变量的非线性估计[J].石油学报,2016,37(12):1543-1549.
XU Baochang,MENG Yu,LIU Wei.Nonlinear estimation of the down-hole unmeasurable variables in the managed pressure drilling system[J].Acta Petrolei Sinica,2016,37(12):1543-1549.
[4] 孙士慧.井底恒压钻井井筒流动模型研究[D].大庆:东北石油大学,2014.
SUN Shihui.Research on wellbore flow model for managed pressure drilling at constant bottomhole pressure[D].Daqing:Northeast Petroleum University,2014.
[5] 王江帅,李军,何岩峰,等.变梯度控压钻井井控过程中井口回压变化规律[J].石油学报,2021,42(11):1499-1505.
WANG Jiangshuai,LI Jun,HE Yanfeng,et al.Variation law of wellhead back pressure under well control during variable gradient managed pressure drilling[J].Acta Petrolei Sinica,2021,42(11):1499-1505.
[6] 徐宝昌,周家立,刘伟,等.基于数据驱动的钻井过程气侵工况预测方法[J].石油学报,2019,40(10):1263-1269.
XU Baochang,ZHOU Jiali,LIU Wei,et al.Data driven prediction method for gas cut in drilling process[J].Acta Petrolei Sinica,2019,40(10):1263-1269.
[7] 宋闯,张晓诚,谢涛,等.渤海"三高"气井环空早期圈闭压力预测[J].石油学报,2022,43(5):694-707.
SONG Chuang,ZHANG Xiaocheng,XIE Tao,et al.Prediction of early annular trap pressure of three-high gas wells in Bohai Sea[J].Acta Petrolei Sinica,2022,43(5):694-707.
[8] 何淼,柳贡慧,李军,等.多相流全瞬态温度压力场耦合模型求解及分析[J].石油钻探技术,2015,43(2):25-32.
HE Miao,LIU Gonghui,LI Jun,et al.Solution and analysis of fully transient temperature and pressure coupling model for multiphase flow[J]. Petroleum Drilling Techniques,2015,43(2):25-32.
[9] 王江帅,李军,柳贡慧,等.变压力梯度下钻井环空压力预测[J].石油学报,2020,41(4):497-504.
WANG Jiangshuai,LI Jun,LIU Gonghui,et al.Prediction of annulus pressure in variable pressure gradients drilling[J].Acta Petrolei Sinica,2020,41(4):497-504.
[10] LAGARIS I E,LIKAS A C,PAPAGEORGIOU D G.Neural-network methods for boundary value problems with irregular boundaries[J].IEEE Transactions on Neural Networks,2000,11(5):1041-1049.
[11] YANG Liu,ZHANG Dongkun,KARNIADAKIS G E.Physics-informed generative adversarial networks for stochastic differential equations[J].SIAM Journal on Scientific Computing,2020,42(1):A292-A317.
[12] LI Hang.Deep learning for natural language processing:advantages and challenges[J].National Science Review,2018,5(1):24-26.
[13] KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[J].Communications of the ACM,2017,60(6):84-90.
[14] ALIPANAHI B,DELONG A,WEIRAUCH M T,et al.Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning[J].Nature Biotechnology,2015,33(8):831-838.
[15] 李宁,徐彬森,武宏亮,等.人工智能在测井地层评价中的应用现状及前景[J].石油学报,2021,42(4):508-522.
LI Ning,XU Binsen,WU Hongliang,et al.Application status and prospects of artificial intelligence in well logging and Formation evaluation[J].Acta Petrolei Sinica,2021,42(4):508-522.
[16] 屈俊波.井底恒压法控压钻井波动压力计算与气侵控制研究[D].大庆:东北石油大学,2019.
QU Junbo.Research on calculation of surge and swab pressures and control of gas kick in constant bottom hole managed pressure drilling[D].Daqing:Northeast Petroleum University,2019.
[17] 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.
[18] MENG Xuhui,LI Zhen,ZHANG Dongkun,et al.PPINN:parareal physics-informed neural network for time-dependent PDEs[J].Computer Methods in Applied Mechanics and Engineering,2020,370:113250.
[19] PANG Guofei,LU Lu,KARNIADAKIS G E.fPINNs:fractional physics-informed neural networks[J].SIAM Journal on Scientific Computing,2019,41(4):A2603-A2626.
[20] YANG Liu,MENG Xuhui,KARNIADAKIS G E.B-PINNs:Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data[J].Journal of Computational Physics,2021,425:109913.
[21] WIGHT C L,ZHAO Jia.Solving Allen-Cahn and Cahn-Hilliard equations using the adaptive physics informed neural networks[J].Communications in Computational Physics,2021,29(3):930-954.
[22] BASSENNE M,FU Lin,MANI A.Time-accurate and highly-stable explicit operators for stiff differential equations[J].Journal of Computational Physics,2021,424:109847.
[23] MCCLENNY L D,BRAGA-NETO U M.Self-adaptive physics-informed neural networks using a soft attention mechanism[C]//Proceedings of the AAAI 2021 Spring Symposium on Combining Artificial Intelligence and Machine Learning with Physical Sciences.Stanford,USA:CEUR-WS.org,2021.
[24] LORDEJANI S N,ABBASI M H,VELMURUGAN N,et al.Modeling and numerical implementation of managed-pressure-drilling systems for the assessment of pressure-control systems[J].SPE Drilling & Completion,2020,35(4):598-619.
[25] LAGE A C V M,TIME R W.Mechanistic model for upward two-phase flow in annuli[R].SPE 63127,2000.
[26] HASAN A R,KABIR C S.Study of multiphase flow behavior in vertical wells[J].SPE Production Engineering,1988,3(2):263-272.
[27] CAETANO F E.Upward vertical two-phase flow through an annulus[J].Journal of Energy Resources Technology,1986,114(1):14-30.
[28] 何淼.控压钻井溢流实时解释理论与控制方法研究[D].北京:中国石油大学(北京),2016.
HE Miao.Influx real time interpretation theory and control method in managed pressure drilling[D].Beijing:China University of Petroleum,2016.
[29] LU Lu,MENG Xuhui,MAO Zhiping,et al.DeepXDE:a deep learning library for solving differential equations[J].SIAM Review,2021,63(1):208-228.
[30] WANG Sifan,YU Xinling,PERDIKARIS P.When and why PINNs fail to train:a neural tangent kernel perspective[J].Journal of Computational Physics,2022,449:110768.
[31] Imran K M J,Amelia R I,Syed Q N.Adam optimization algorithm for wide and deep neural network[J].Knowledge Engineering and Data Science,2019,2(1):41-46.
[32] LIU D C,NOCEDAL J.On the limited memory BFGS method for large scale optimization[J].Mathematical Programming,1989,45(1-3):503-528.

用于两相流环空压力预测的自适应物理信息神经网络模型

Self-adaptive physical information neural network model for prediction of two-phase flow annulus pressure

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

本文评价

[1]	路保平. 高酸性活跃厚沥青层复杂碳酸盐岩油田钻完井关键技术[J]. 石油学报, 2021, 42(7): 962-970.
[2]	许林, 刘书杰, 许明标, 冯桓榰, 邢希金, 邓佳佳. 压差激活密封剂的微缺陷自适应修复行为及机理[J]. 石油学报, 2021, 42(5): 686-694.
[3]	暴丹, 邱正松, 叶链, 钟汉毅, 赵欣, 邱维清, 王宝田, 郭保雨. 热致形状记忆“智能”型堵漏剂的制备与特性实验[J]. 石油学报, 2020, 41(1): 106-115.
[4]	糜利栋, 姜汉桥, 胡向阳, 李俊键, 胡小虎, 周元龙, 贾英, 严必成. 复杂裂缝网络页岩气藏自适应网格剖分方法[J]. 石油学报, 2019, 40(2): 197-206.
[5]	唐杰, 张文征, 温雷, 谷玉田, 陈学国. 基于正则化局部自适应核回归的单点高密度地震资料随机噪声压制方法[J]. 石油学报, 2019, 40(12): 1495-1502,1552.
[6]	徐宝昌, 周家立, 刘伟, 付加胜. 基于数据驱动的钻井过程气侵工况预测方法[J]. 石油学报, 2019, 40(10): 1263-1269.
[7]	程浩, 陈刚, 王恩德, 侯振隆, 付建飞. 基于Shearlet变换的自适应阈值地震数据去噪方法[J]. 石油学报, 2018, 39(1): 82-91.
[8]	徐春露, 孙建孟, 董旭, 赖富强. 页岩气储层孔隙压力测井预测新方法[J]. 石油学报, 2017, 38(6): 666-676.
[9]	沈跃, 张亨, 张令坦, 苏义脑, 盛利民, 李林. 基于自适应滤波的钻井液连续压力波信号噪声抑制[J]. 石油学报, 2014, 35(2): 353-358.
[10]	李琳努尔扎提张奇志. 钻机旋转系统的自适应PID控制策略[J]. 石油学报, 2011, 32(1): 158-161.
[11]	朱晓东王杰. 基于分层模糊系统的石油钻井参数预测模型[J]. 石油学报, 2010, 31(5): 838-842.
[12]	易方, 李著信, 苏毅, 王鹏飞, 吴昊. 基于改进型小波阈值的输油管道磁记忆信号降噪方法[J]. 石油学报, 2009, 30(1): 141-144.
[13]	吕开河, 邱正松, 魏慧明, 李先锋, 宋元森. 自适应防漏堵漏钻井液技术研究[J]. 石油学报, 2008, 29(5): 757-760,765.
[14]	伦淑娴, 张化光, 冯健. 自适应模糊神经网络系统在管道泄漏检测中的应用[J]. 石油学报, 2004, 25(4): 101-104.
[15]	白莉, 岳前进, 李洪升. 噪声干扰条件下长输管道检漏的一致最大功效检验[J]. 石油学报, 2004, 25(2): 108-112.