基于条件生成式对抗网络的油藏单井产量预测模型

doi:10.19596/j.cnki.1001-246x.8480

摘要/Abstract

摘要：

为克服机器学习方法在油藏单井产量预测中的过拟合问题, 提高油田生产中的产量预测精度, 提出一种基于条件生成式对抗网络(CGAN)的油藏单井产量预测模型。该模型使用长短期记忆、全连接等基础神经网络, 构建生成和判别网络模型。生成网络模型以产量影响因素为条件输入, 生成预测产量数据, 利用对数损失函数评价预测数据与真实数据之间的偏差, 通过条件生成式对抗网络的博弈训练, 并结合贝叶斯超参数优化算法, 优化模型结构, 综合提高模型的泛化能力。基于Eclipse数值模拟软件建立同一井网条件下不同地质和生产条件下的油藏单井产量数据库, 以地质与生产条件等产量影响因素作为模型的条件输入, 进行油藏单井产量预测。结果表明: 与全连接神经网络(FCNN)、随机森林(RF)以及长短期记忆神经网络(LSTM)模型的预测结果相比, CGAN模型在测试集上的平均绝对百分比误差分别提升了2.59%、0.81%以及1.72%, 并且过拟合比最小(1.027)。说明CGAN降低了机器学习产量预测模型的过拟合程度, 提高了模型的泛化能力与预测精度, 验证了所提算法的优越性, 对指导油田高效开发和保障我国能源战略安全具有重要意义。

关键词: 条件生成式对抗网络, 产量预测, 机器学习, 贝叶斯超参数预测, 神经网络

Abstract:

To address overfitting problem of production forecast model in machine learning and improve accuracy of production forecast in actual oil field, a model for single well production forecasting of reservoir based on Conditional Generative Adversarial Net(CGAN) was proposed. The model hybridizes two types of basic neural network structures, i.e., long short-term memory network and fully connected network, to construct a generative model and a discriminant model. The generative model takes production influencing factors as conditions to generate the forecasting production data. It defines a logarithmic loss function as a residual between predicted and real data, to improve comprehensively the generalization ability of the model. Bayesian hyperparameter optimization algorithm was used to optimize the model structure through game training of CGAN. With numerical simulation software Eclipse, single well production database with same well pattern under different geological and production conditions was established to train CGAN, which can be used to predict rapidly single well production of reservoir by taking geological and production factors as condition input of the model. Experimental results show that compared with prediction results of models FCNN, RF, and LSTM, mean absolute percentage error of CGAN model on the test set is increased by 2.59%, 0.81%, and 1.72%. The overfitting ratio is the smallest(1.027). It indicates that CGAN reduces overfitting degree of the machine-learning-based production forecast model, and improves generalization ability and accuracy of the model as well. This verifies superiority of the algorithm, which is of great significance to guide efficient development of oilfields and ensures security of national energy strategy.

Key words: conditional generative adversarial net, production forecast, machine learning, Bayes hyperparameter optimization, neural network

黄灿, 田冷, 王恒力, 王嘉新, 蒋丽丽. 基于条件生成式对抗网络的油藏单井产量预测模型[J]. 计算物理, 2022, 39(4): 465-478.

Can HUANG, Leng TIAN, Heng-li WANG, Jia-xin WANG, Li-li JIANG. A Single Well Production Forecasting Model of Reservoir Based on Conditional Generative Adversarial Net[J]. Chinese Journal of Computational Physics, 2022, 39(4): 465-478.

图/表 16

图1 GAN基本结构

Fig.1 GAN basic structure

图2 CGAN基本结构

Fig.2 CGAN basic structure

图3 基于CGAN的单井产量预测模型

Fig.3 Single well production prediction model based on CGAN

图4 生成网络的基本结构

Fig.4 Generation of basic structure of the network

图5 判别网络的基本结构

Fig.5 Identification of basic structure of the network

图6 训练判别网络的算法描述流程示意图

Fig.6 Schematic of algorithm description process for training the discriminant network

图7 训练生成网络的算法描述流程示意图

Fig.7 Schematic of algorithm description process for training the generative network

图8 油藏注采井网示意图

Fig.8 Schematic of reservoir injection-production well pattern

图9 平均绝对百分比误差变化

Fig.9 Mean absolute percentage error

图10 生成网络与判别网络误差变化

Fig.10 Logarithmic error of generative network and discriminant network

图11 测试集中预测产量与实际产量(井口1~12)

Fig.11 Predicted output and actual output of test sets (Y1-Y12)

图12 测试集中预测产量与实际产量(井口13~24)

Fig.12 Predicted output and actual output of test sets (Y13-24)

表1 产量预测平均绝对百分比误差

Table 1 Average absolute percentage errors of yield forecast

模型	平均绝对百分比误差/%		过拟合比
模型	验证集	测试集	过拟合比
CGAN	4.68	7.20	1.027
FCNN	4.46	9.79	1.059
LSTM	2.16	8.01	1.064
RF	5.90	8.92	1.033

图13 Q1油井2年内的产量随S1区域初始地层渗透率的变化

Fig.13 2-year production of well Q1 as functions of initial formation permeability in S1 area

图14 Q1油井2年内的产量随S1区域初始地层孔隙度的变化

Fig.14 2-year production of well Q1 as functions of initial formation porosity in S1 area

图15 Q1油井2年内的产量随注水井W1的月注水量的变化

Fig.15 Production of well Q1 in 2 years as functions of monthly water injection of well W1

参考文献 26

1	MALE F, AIKEN C, DUNCAN I J. Using data analytics to assess the impact of technology change on production forecasting[C]//Proceedings of the 2018 Society of Petroleum Engineers Conference on Annual Technical, 2018: 1-14.
2	ZHANG Y , HU J H , ZHANG Q . Application of locality preserving projection-based unsupervised learning in predicting the oil production for low-permeability reservoirs[J]. SPE Journal, 2021, 26 (3): 1302- 1313. DOI
3	JIA P , CHEN L S , HUANG S J , et al. Characterization of fracture network by volume fracturing in horizontal wells and coupled flow model[J]. Chinese Journal of Computational Physics, 2015, 32 (06): 685- 692. DOI
4	ZHANG J , HUANG S J , CHEN L S . Monte carlo Calculation of stable productivity of fractured directional wells in natural fracture reservoirs[J]. Chinese Journal of Computational Physics, 2014, 31 (05): 567- 572. DOI
5	BOOMER R J. Predicting production using a neural network (artificial intelligence beats human intelligence)[C]//Proceedings of the 1995 Society of Petroleum Engineers Conference on Petroleum Computer, 1995: 195-204.
6	吴新根, 葛家理. 应用人工神经网络预测油田产量[J]. 石油勘探与开发, 1994, (03): 75- 78.
7	NGUYEN H H , CHAN C W , WILSON M . Prediction of oil well production: A multiple-neural-network approach[J]. Intelligent Data Analysis, 2004, 8 (2): 183- 196. DOI
8	BANSAL Y, ERTEKIN T, KARPYN Z, et al. Forecasting well performance in a discontinuous tight oil reservoir using artificial neural networks[C]//Proceedings of the 2013 Society of Petroleum Engineers Conference on Unconventional Resources, 2013: 1-12.
9	CAO Q, BANERJEE R, GUPTA S, et al. Data driven production forecasting using machine leaning[C]//Proceedings of the 2016 Society of Petroleum Engineers Conference on Argentina Exploration and Production of Unconventional Resources Symposium, 2016: 1-10.
10	WANG S, CHEN S. A comprehensive evaluation of well completion and production performance in bakken shale using data-driven approaches[C]//Proceedings of the 2016 Society of Petroleum Engineers Conference on Asia Pacific Hydraulic Fracturing, 2016: 1-25.
11	JEONG H , SUN A Y , LEE J , et al. A learning-based data-driven forecast approach for predicting future reservoir performance[J]. Advances in Water Resources, 2018, 118, 95- 109. DOI
12	SUN J, MA X, KAZI M. Comparison of decline curve analysis DCA with recursive neural networks RNN for production forecast of multiple wells[C]//Proceedings of the 2018 Society of Petroleum Engineers Conference on Western Regional Meeting, 2018: 1-11.
13	ZHOU Y H , LIU H Q , QI P , et al. Forecast of oil production in fractured-vuggy reservoir by using recurrent neural networks[J]. Chinese Journal of Computational Physics, 2018, 35 (06): 668- 674.
14	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 (05): 1009- 1015.
15	BHATTACHARYA S , GHAHFAROKHI P K , CARR T R , et al. Application of predictive data analytics to model daily hydrocarbon production using petrophysical, geomechanical, fiber-optic, completions, and surface data: A case study from the marcellus shale, North America[J]. Journal of Petroleum Science and Engineering, 2019, 176, 702- 715. DOI
16	LIAO L, ZENG Y, LIANG Y, et al. Data mining: A novel strategy for production forecast in tight hydrocarbon resource in Canada by random forest analysis[C]//Proceedings of the 2020 International Conference on Petroleum Technology, 2020: 1-11.
17	陈森朋, 吴佳, 陈修云. 基于强化学习的超参数优化方法[J]. 小型微型计算机系统, 2020, 41 (04): 679- 684.
18	邓帅. 基于改进贝叶斯优化算法的CNN超参数优化方法[J]. 计算机应用研究, 2019, 36 (07): 1984- 1987.
19	GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems, 2014, 2: 2672-2680.
20	MIRZA M , OSINDERO S . Conditional generative adversarial nets[J]. Computer Science, 2014, 2672- 2680.
21	林珊, 王红, 齐林海, 等. 基于条件生成对抗网络的短期负荷预测[J]. 电力系统自动化, 2021, 45 (11): 52- 60.
22	蒋华伟, 张磊. 基于长短期记忆生成对抗网络的小麦品质多指标预测模型[J]. 电子与信息学报, 2020, 42 (12): 2865- 2872.
23	NASH J F . Equilibrium points in n-person games[J]. Proceedings of the National Academy of Sciences, 1950, 36 (1): 48- 49.
24	MASKIN E . Nash equilibrium and welfare optimality[J]. The Review of Economic Studies, 1999, 66 (1): 23- 38.
25	崔佳旭, 杨博. 贝叶斯优化方法和应用综述[J]. 软件学报, 2018, 29 (10): 3068- 3090.
26	陈文兵, 管正雄, 陈允杰. 基于条件生成式对抗网络的数据增强方法[J]. 计算机应用, 2018, 38 (11): 3305- 3311.

[1]	吴国正, 王发杰, 程隋福, 张成鑫. 基于物理信息神经网络的内部声场正反问题数值计算[J]. 计算物理, 2022, 39(6): 687-698.
[2]	肖聪, 张士诚, 马新仿, 周彤, 侯腾飞. 基于模型降维和递归神经网络的油藏参数反演[J]. 计算物理, 2022, 39(5): 564-578.
[3]	潘剑, 郭照立, 陈松泽. 从噪声数据学习偏微分方程的复合神经网络[J]. 计算物理, 2022, 39(2): 223-232.
[4]	唐利红, 贺宗梅, 姚延立. 忆阻Hopfield神经网络动力学分析及其电路实现[J]. 计算物理, 2022, 39(2): 244-252.
[5]	李洪, 章立新, 任燕, 高明, 刘婧楠. 基于灰色关联分析的BP神经网络对混流闭式冷却塔出水温度的预测[J]. 计算物理, 2022, 39(1): 53-59.
[6]	罗佳, 孙亮, 乔印虎. 忆阻突触耦合环形Hopfield神经网络动力学分析及其电路实现[J]. 计算物理, 2022, 39(1): 109-117.
[7]	郑书昱, 彭佳臻, 张先梅, 薛二兵, 虞立敏. 基于神经网络的托卡马克能量约束时间预测[J]. 计算物理, 2021, 38(4): 423-430.
[8]	周倩, 韦笃取. 场耦合忆阻神经网络的电活动[J]. 计算物理, 2020, 37(6): 750-756.
[9]	卢新瑞, 黄捍东, 李帅, 尹龙. 基于U-Net的盐体识别方法[J]. 计算物理, 2020, 37(3): 327-334.
[10]	周于皓, 刘慧卿, 祁鹏, 赵萌, 陈宇. 基于循环神经网络的缝洞型油藏油井产量预测[J]. 计算物理, 2018, 35(6): 668-674.
[11]	苟雪银, 郭立新, 张连波. 支持向量机和神经网络在粗糙面参数反演中的比较[J]. 计算物理, 2014, 31(1): 75-84.
[12]	李旭, 俞集辉, 李永明, 汪泉弟. 互连导线串扰问题的人工神经网络预测[J]. 计算物理, 2009, 26(2): 311-316.
[13]	孟娟, 王兴元. 一类延迟混沌神经网络的鲁棒反同步[J]. 计算物理, 2008, 25(2): 247-252.
[14]	王兴元, 谭贵霖. Liley模型的模拟EEG信号的非线性预测和分析[J]. 计算物理, 2007, 24(5): 612-618.
[15]	张全虎, 张其欣, 吕峰, 李泽, 李鲲鹏, 王仲奇, 赵学军, 隋洪志. 基于径向基函数神经网络的线性衰减系数的重建算法[J]. 计算物理, 2003, 20(5): 439-442.