A methodology for predicting production dynamics of horizontal wells in bottom water reservoirs by integrating data-driven approaches with percolation mechanisms

doi:10.3969/j.issn.2096-1693.2025.03.020

Abstract

Abstract:

The production dynamics of horizontal wells in bottom-water sandstone reservoirs are influenced by multiple factors, including reservoir properties and production regimes, posing significant challenges for prediction. Traditional empirical formulas and numerical simulation methods have numerous limitations. In recent years, with the advancement of intelligent oilfield processes, machine learning approaches have been increasingly applied to oilfield production processes. Compared to traditional methods, machine learning offers comprehensiveness and efficiency but lacks physical interpretability and robustness, thereby restricting its credibility and applicability in practical engineering scenarios. To address this issue, firstly, taking a block in the Bohai Oilfield as an example, this paper establishes a seepage mechanism model for horizontal wells in bottom-water sandstone reservoirs considering time-varying permeability. Secondly, improvements are made to the traditional Long Short-Term Memory (LSTM) neural network structure to construct a neural network framework that supports multi-layer inputs of both static and dynamic data, enhancing the model’s adaptability to various geological conditions, with adjustable weights for static and dynamic parameters. Lastly, a dataset generated from the seepage mechanism model is used, and through dataset fusion, a hybrid prediction model combining data-driven and seepage physics mechanisms for horizontal well production dynamics is established. This data-physics fusion mechanism enhances the interpretability and robustness of the model. Test results indicate that the multi-layer input neural network framework for static and dynamic parameters achieves slightly higher prediction accuracy compared to traditional LSTM networks, while the hybrid model integrating data-driven and seepage mechanism shows a significantly improved prediction accuracy over the multi-layer input neural network framework. Field application also verifies the high efficiency and practicality of this hybrid model in predicting the production dynamics of horizontal wells, providing strong support for optimizing oilfield development plans and establishing production regimes.

Key words: horizontal well production dynamic prediction, Long Short-Term Memory neural network, dynamic and static combination, physical mechanism, integrated model

摘要：

底水砂岩油藏水平井生产动态受储层属性、生产制度等多因素影响，预测难度大。传统经验公式和数值模拟等方法存在诸多局限性。近年来，伴随油田智能化进程，机器学习方法逐渐广泛应用于油田生产过程。相对于传统方法，该方法兼具全面性与高效性，但缺乏物理可解释性和鲁棒性，限制了其在实际工程中的可信度与适用性。为解决该问题，首先，本文以渤海油田某区块为例，建立考虑渗透率时变的底水砂岩油藏水平井渗流机理模型。其次，本文对传统LSTM神经网络结构进行了改进，构建了一个支持动静态数据多层输入的神经网络框架，提升了模型对不同地质状况的适应性，且该框架的动静态参数权重可调整。最后，本文用渗流机理模型生成机理模型数据集，并采用数据集融合方式建立了数据驱动与渗流物理机理融合的水平井生产动态预测模型，数据—物理融合机制提升了该模型的可解释性和鲁棒性。测试结果表明：动静态参数多层输入神经网络框架相对于传统LSTM神经网络预测精度小幅提升；数据驱动与渗流机理融合模型相比于动静态参数多层输入神经网络框架预测精度显著提升。现场应用效果也验证了该融合模型在水平井生产动态预测中的高效性和实用性，为油田开发方案优化和生产制度制定提供了有力支撑。

关键词: 水平井生产动态预测, 长短期记忆神经网络, 动静结合, 物理机理, 融合模型

CLC Number:

TE312
TP18

JIN Qingshuang, XUE Yongchao, ZHANG Ying, LIU Xiaoqi, LI Quan, ZHENG Aile. A methodology for predicting production dynamics of horizontal wells in bottom water reservoirs by integrating data-driven approaches with percolation mechanisms[J]. Petroleum Science Bulletin, 2025, 10(5): 997-1014.

金青爽, 薛永超, 张颖, 刘晓旗, 李权, 郑爱乐. 数据驱动与渗流机理融合的底水油藏水平井生产动态预测方法[J]. 石油科学通报, 2025, 10(5): 997-1014.

/ / Recommend / Download Citations

URL: http://sykxtb.dgqinyehang.com/EN/10.3969/j.issn.2096-1693.2025.03.020

http://sykxtb.dgqinyehang.com/EN/Y2025/V10/I5/997

Figures/Tables 24

Fig. 1 Physical model

Fig. 2 Discrete process of the numerical model

Fig. 3 Model solution process

Fig. 4 Architecture of neural network with dynamic and static combined parameter input

Fig. 5 Data preparation process

Fig. 6 Integration method of data-driven and seepage mechanism

Fig. 7 Optimization of static parameters

Fig. 8 Optimization of dynamic parameters

Fig. 9 Dynamic data formatting processing

Fig. 10 Schematic diagram of dataset relationship

Fig. 11 Flowchart of neural network architecture optimization

Fig. 12 The effect of improving the accuracy of the dynamic and static combined model

Table 1 Prediction error evaluation table

模型	含水率		井底流压		产液量
模型	绝对误差/%	相对误差/%	绝对误差/MPa	相对误差/%	绝对误差/(m³/d)	相对误差/%
动态数据驱动模型	5.6	18.43	0.78	16.35	25.1	15.97
动静结合数据驱动模型	4.4	16.21	0.66	15.32	20.3	13.76

Fig. 13 Distribution range of each attribute

Table 2 The results (part) of the orthogonal experimental design

井名	渗透率/mD	孔隙度/%	含水饱和度/%	有效厚度/m	底水厚度/m
H1	1500	26	25	16	8.0
H2	1722	27	27	16	10.2
H3	1944	28	29	16	6.4
H4	2166	29	32	16	4.9
H5	2388	3	34	16	11.8

Fig. 14 Oil saturation fields at different times (considering the time-varying permeability)

Fig. 15 Oil saturation profile after 20 years of production

Fig. 16 Changes of daily oil production and water cut under time-varying permeability

Fig. 17 Evaluation of absolute error under the proportion of different mechanism model data

Fig. 18 The average relative error changes of liquid produc-tion, water cut and bottom hole pressure

Fig. 19 Variation of the determination coefficient of the prediction results with the proportion of the mechanism model

Fig. 20 Evaluation results of the test set when the proportion of the mechanism model is 40%

Fig. 21 Evaluation of the prediction effect of well X15

Table 3 Evaluation table of relative errors predicted by each model

预测模型	预测项
预测模型	产液量/%	含水率/%	井底流压/%
数据驱动模型(仅考虑动态参数)	15.97	18.43	16.35
动静结合数据驱动模型	13.76	16.21	15.32
数据与渗流机理融合模型	7.43	5.33	12.33

References 43

[1]	喻高明, 凌建军, 蒋明煊, 等. 砂岩底水油藏开采机理及开发策略[J]. 石油学报, 1997, 18(2): 64-68.
	[YU G M, LING J J, JIANG M X, et al. Exploitation mechanism and development strategy of sandstone bottom water reservoirs[J]. Acta Petrolei Sinica, 1997, 18(2): 64-68.]
[2]	喻高明, 凌建军, 蒋明煊. 砂岩底水油藏底水锥进影响因素研究[J]. 江汉石油学院学报, 1996, (3): 59-62.
	[YU G M, LING J J, JIANG M X. Research on influencing factors of bottom water cone advance in sandstone bottom water reservoirs[J]. Journal of Jianghan Petroleum Institute, 1996, (3): 59-62.]
[3]	邹威, 姚约东, 王庆, 等. 底水油藏水平井水脊形态影响因素[J]. 油气地质与采收率, 2017, 24(5): 70-77.
	[ZOU W, YAO Y D, WANG Q, et al. Horizontal wells in bottom water reservoir ridge form factors[J]. Geology and oil recovery of oil and gas, 2017, 24(5): 70-77.]
[4]	刘学利, 郑小杰, 吴居林, 等. 塔河强底水砂岩油藏水驱曲线对比分析及稳定水驱段识别[J]. 复杂油气藏, 2024, 17(4): 417-425.
	[LIU X L, ZHENG X J, WU J L, et al. Tahe of bottom water sandstone oil reservoir water drive curve contrast analysis and the stability of water flooding period of recognition[J]. Journal of complex reservoirs, 2024, 17(4): 417-425.]
[5]	刘学利, 郑小杰, 窦莲, 等. 薄层强底水多韵律层砂岩油藏高精细数值模拟研究——以塔河9区下油组油藏为例[J]. 油气藏评价与开发, 2022, 12(2): 391-398.
	[LIU X L, ZHENG X J, DOU L, et al. Research on high-precision numerical simulation of thin-layer strong bottom water multi--rhythm layer sandstone reservoirs: a case study of the Xiayou group reservoir in area 9 of Tahe[J]. Reservoir evaluation and development, 2022, 12(2): 391-398.]
[6]	张凯, 赵兴刚, 张黎明, 等. 智能油田开发中的大数据及智能优化理论和方法研究现状及展望[J]. 中国免费靠逼视频学报(自然科学版), 2020, 44(4): 28-38.
	[ZHANG K, ZHAO X G, ZHANG L M, et al. Research status and prospect of big data and intelligent optimization theory and methods in intelligent oilfield development[J] Journal of China University of Petroleum (Natural Science Edition), 2020, 44(4): 28-38.]
[7]	匡立春, 刘合, 任义丽, 等. 人工智能在石油勘探开发领域的应用现状与发展趋势[J]. 石油勘探与开发, 2021, 48(1): 1-11. doi: 10.11698/PED.2021.01.01
	[KUANG L C, LIU H, REN Y L, et al. The application status and development trend of artificial intelligence in the field of petroleum exploration and development[J]. Petroleum Exploration and Development, 2021, 48(1): 1-11.]
[8]	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. doi: 10.7623/syxb202104008
	[LI N, XU B S, WU H L, et al. The current situation and prospect of the application of artificial intelligence in logging formation evaluation[J]. Acta Petrolei Sinica, 2021, 42(4): 508-522.]
[9]	赵改善. 石油物探智能化发展之路: 从自动化到智能化[J]. 石油物探, 2019, 58(6): 791-810. doi: 10.3969/j.issn.1000-1441.2019.06.002
	[ZHAO G S. The intelligent development path of petroleum geophysical exploration: from automation to intelligence[J]. Petroleum Geophysical Exploration, 2019, 58(6): 791-810.]
[10]	肖立志. 机器学习数据驱动与机理模型融合及可解释性问题[J]. 石油物探, 2022, 61(2): 205-212. doi: 10.3969/j.issn.1000-1441.2022.02.002
	[XIAO L Z. The fusion and interpretability of machine learning data-driven and mechanism models[J]. Geophysical Prospecting for Petroleum, 2022, 61(2): 205-212.] doi: 10.3969/j.issn.1000-1441.2022.02.002
[11]	闵超, 代博仁, 张馨慧, 等. 机器学习在油气行业中的应用进展综述[J]. 西南免费靠逼视频学报(自然科学版), 2020, 42(6): 1-15. doi: 10.11885/j.issn.1674-5086.2020.06.05.03
	[MIN C, DAI B R, ZHANG X H, et al. Review of the application progress of machine learning in the oil and gas industry[J]. Journal of Southwest Petroleum University (Natural Science Edition), 2020, 42(6): 1-15.]
[12]	杨红梅, 薛敏, 杨泱, 等. 基于机器学习的页岩油藏合理焖井时间预测[J]. 西安免费靠逼视频学报(自然科学版), 2022, 37(2): 65-72.
	[YANG H M, XUE M, YANG Y, et al. Prediction of reasonable well trapping time in shale oil reservoirs based on machine learning[J]. Journal of Xi ‘an Shiyou University (Natural Science Edition), 2022, 37(2): 65-72.]
[13]	胡晓东, 涂志勇, 罗英浩, 等. 拟合函数—神经网络协同的页岩气井产能预测模型[J]. 石油科学通报, 2022, 7(3): 394-405.
	[HU X D, TU Z Y, LUO Y H, et al. Shale gas well productivity prediction model based on fitting function-neural network collaboration[J]. Bulletin of Petroleum Science, 2022, 7(3): 394-405.]
[14]	杨二龙, 陈柄君, 董驰, 等. 一种基于机器学习的井间水驱优势通道识别方法[J]. 钻采工艺, 2025, 48(1): 157-164.
	[YANG E L, CHEN B J, DONG C, et al. A machine learning-based method for identifying advantageous channels in inter-well water flooding[J]. Drilling & Production Technology, 2025, 48(1): 157-164.]
[15]	孙予舒, 黄芸, 梁婷, 等. 基于XGBoost算法的复杂碳酸盐岩岩性测井识别[J]. 岩性油气藏, 2020, 32(4): 98-106. doi: 10.12108/yxyqc.20200410
	[SUN Y S, HUANG Y, LIANG T, et al. Identification of lithology logging of complex carbonate rocks based on XGBoost algorithm[J]. Lithologic Reservoirs, 2020, 32(4): 98-106.]
[16]	石兰香, 苟燕, 李秀峦, 等. 稠油油藏双水平井SAGD蒸汽腔上升阶段产量预测解析模型[J]. 石油科学通报, 2022, 7(1): 106-115.
	[SHI L X, GOU Y, LI X L, et al. Analytical model for production prediction in the ascending stage of SAGD steam chambers in double horizontal wells of heavy oil reservoirs[J]. Bulletin of Petroleum Science, 2022, 7(1): 106-115.]
[17]	NEGASH B M, YAW A D. 基于人工神经网络的注水开发油藏产量预测[J]. 石油勘探与开发, 2020, 47(2): 357-365. doi: 10.11698/PED.2020.02.14
	[NEGASH B M, YAW A D. Artificial neural network based production forecasting for a hydrocarbon reservoir under water injection[J]. Petroleum Exploration and Development, 2020, 47(2): 357-365.]
[18]	曹冲, 程林松, 张向阳, 等. 基于多变量小样本的渗流代理模型及产量预测方法[J]. 力学学报, 2021, 53(8): 2345-2354.
	[CAO C, CHENG L S, ZHANG X Y, et al. Seepage agent model and production prediction method based on multivariate small samples[J]. Acta Mechanica Sinica, 2021, 53(8): 2345-2354.]
[19]	薛永超, 袁志乾, 金青爽, 等. 基于深度森林算法的油井产量预测[J]. 科学技术与工程, 2022, 22(11): 4327-4334.
	[XUE Y C, YUAN Z Q, JIN Q S, et al. Oil well production prediction based on deep forest algorithm[J]. Science Technology and Engineering, 2022, 22(11): 4327-4334.]
[20]	谷建伟, 周梅, 李志涛, 等. 基于数据挖掘的长短期记忆网络模型油井产量预测方法[J]. 特种油气藏, 2019, 26(2): 77-81.
	[GU J W, ZHOU M, LI Z T, et al. Oil well production prediction method based on Long Short-Term Memory network model of data mining[J]. Special Oil & Gas Reservoirs, 2019, 26(2): 77-81.]
[21]	王洪亮, 穆龙新, 时付更, 等. 基于循环神经网络的油田特高含水期产量预测方法[J]. 石油勘探与开发, 2020, 47(5): 1009-1015. doi: 10.11698/PED.2020.05.15
	[WANG H L, MU L X, SHI F G, et al. Production prediction method for oilfield during extremely high water cut period based on recurrent neural network[J]. Petroleum Exploration and Development, 2020, 47(5): 1009-1015.]
[22]	潘少伟, 郑泽晨, 王吉哲, 等. 基于长短期记忆网络和注意力机制的油井产油量预测[J]. 科学技术与工程, 2021, 21(30): 13010-13015.
	[PAN S W, ZHENG Z C, WANG J Z, et al. Oil well oil production prediction based on Long Short-Term Memory network and attention mechanism[J]. Science Technology and Engineering, 2021, 21(30): 13010-13015.]
[23]	谢子琳. 基于改进粒子群算法和长短期记忆神经网络的油井产量预测模型[C]// 中国石油新疆油田分公司(新疆砾岩油藏实验室), 西安免费靠逼视频, 陕西省石油学会. 2022油气田勘探与开发国际会议论文集Ⅲ., 中国免费靠逼视频(华东), 2022: 281-291.
	[XIE Z L. Oil well production prediction model based on improved particle swarm optimization algorithm and Long Short-Term Memory neural network[C]// Xinjiang Oilfield Branch of China National Petroleum Corporation (Xinjiang Conglomerate Reservoir Laboratory), Xi ‘an Shiyou University, Shaanxi Petroleum Society. Proceedings of the 2022 International Conference on Oil and Gas Field Exploration and Development III. China University of Petroleum (East China).]
[24]	李成彬. 基于改进LSTM的复杂断块油藏油井产量预测方法[J]. 石化技术, 2024, 31(4): 258-260.
	[LI C B. Oil well production prediction method for complex fault block reservoirs based on improved LSTM[J] Petrochemical Technology, 2024, 31(4): 258-260.]
[25]	王洪亮, 林霞, 蒋丽维, 等. 基于聚类及长短时记忆神经网络预测油田产量[J]. 石油科学通报, 2024, 9(1): 62-72.
	[WANG H L, LIN X, JIANG L W, et al. Prediction of oilfield production based on clustering and Long Short-Term Memory neural network[J]. Bulletin of Petroleum Science, 2024, 9(1): 62-72.]
[26]	韩江峡, 薛亮, 位云生, 等. 基于深度自回归神经网络的多井产量概率预测[J]. 石油科学通报, 2024, 9(4): 679-689.
	[HAN J X, XUE L, WEI Y S, et al. Probability prediction of multi-well production based on deep autoregressive neural network[J]. Chinese Petroleum Science Bulletin, 2024, 9(4): 679-689.]
[27]	倪莹莹, 唐慧莹, 贺戈, 等. 基于改进的长短期记忆神经网络的页岩油井产量预测[J/OL]. 断块油气田, 1-17. [2025-07-04].
	[NI Y Y, TANG H Y, HE G, et al. Shale oil well production prediction based on improved Long Short-Term Memory neural network[J/OL]. Fault-block oil and gas field, 1-17. [2025-07-04].]
[28]	张晓东, 李敏. 基于WPD-FEEMD和ARIMA-LSTM的油井产量预测方法[J]. 传感器与微系统, 2025, 44(6): 161-164.
	[ZHANG X D, LI M. Oil well production prediction method based on WPD-FEEMD and ARIMA-LSTM[J]. Sensors and Microsystems, 2025, 44(6): 161-164.]
[29]	张瑞, 贾虎. 基于多变量时间序列及向量自回归机器学习模型的水驱油藏产量预测方法[J]. 石油勘探与开发, 2021, 48(1): 175-184. doi: 10.11698/PED.2021.01.16
	[ZHANG R, JIA H. Production prediction method for water-driven reservoirs based on multivariate time series and vector autoregressive machine learning model[J]. Petroleum Exploration and Development, 2021, 48(1): 175-184.]
[30]	毕剑飞, 李靖, 吴克柳, 等. 数据驱动与物理驱动融合的双驱动渗流代理模型构建[J]. 油气地质与采收率, 2023, 30(3): 104-114.
	[BI J F, LI J, WU K L, et al. Data driven and physical drive fusion of the two drive seepage agent model to build[J]. Geology and oil recovery of oil and gas, 2023, 30(3): 104-114.]
[31]	李野, 陈松灿. 基于物理信息的神经网络: 最新进展与展望[J]. 计算机科学, 2022, 49(4): 254-262. doi: 10.11896/jsjkx.210500158
	[LI Y, CHEN S C. Neural networks based on physical information: latest advances and prospects[J]. Computer Science, 2022, 49(4): 254-262.] doi: 10.11896/jsjkx.210500158
[32]	ZENG T, ZENG N, DENG C, et al. A data-physical dual-driven surrogate model for reservoir simulation[J]. Physics of Fluids, 2025, 37(2): 1-11.
[33]	SONG H, DU S, YANG J, et al. Evaluation of hydraulic fracturing effect on coalbed methane reservoir based on deep learning method considering physical constraints[J]. Journal of Petroleum Science and Engineering, 2022, 212: 110360.
[34]	MAQUI A F, ZHAI X, SUAREZ N A, et al. A comprehensive workflow for near real time waterflood management and production optimization using reduced-physics and data-driven technologies[C]// SPE Latin America and Caribbean Petroleum Engineering Conference. SPE, 2017: D021S008R008.
[35]	CHENG S, ALKHALIFAH T. Unit-constrained data-driven discovery of a wave equation[C]// SEG International Exposition and Annual Meeting. SEG, 2024: SEG-2024-4098603.
[36]	沈路航. 基于全连接神经网络的单相渗流方程求解方法研究[D]. 合肥: 合肥工业大学, 2024.
	[SHEN L H. Research on the solution method of single-phase seepage equation based on fully connected neural network[D]. Hefei: Hefei University of Technology, 2024.]
[37]	王昀卓. 求解偏微分方程的神经网络方法[D]. 合肥: 中国科学技术大学, 2021.
	[WANG Y Z. To solve the partial differential equation of neural network method[D]. Hefei: University of science and technology of China, 2021.]
[38]	ALMAJID M M, ABU A M O. Prediction of porous media fluid flow using physics informed neural networks[J]. Journal of Petroleum Science and Engineering, 2021, 208: 1-17
[39]	LI D L, SHEN L H, ZHA W S, et al. Physics-constrained deep learning for solving seepage equation[J]. Journal of Petroleum Science and Engineering, 2021, 206: 1-11.
[40]	PU J, SONG W F, Wu J L, et al. PINN-based method for predicting flow field distribution of the tight reservoir after fracturing[J]. GEOFLUIDS, 2022, 2022(1): 2-10.
[41]	WANG H, WANG M M, CHEN S N, et al. A novel governing equation for shale gas production prediction via physics-informed neural networks[J]. Expert Systems with Applications: An International Journal, 2024, 248(C): 1-11.
[42]	GUO H, WU S. Solution to the two-phase flow in heterogeneous porous media based on physics-informed neural network[J]. Chemistry & Technology of Fuels & Oils, 2024, 60(5): 1188-1196.
[43]	LIU B T, WEI J, KANG L X, et al. Physics-informed neural network (PINNs) for convection equations in polymer flooding reservoirs[J]. PHYSICS OF FLUIDS, 2025, 37(3): 1-10.

Please choose a citation manager

Content to export