基于参数优选的储层渗透率深度置信网络模型预测初探

doi:10.13809/j.cnki.cn32-1825/te.2021.04.014

摘要/Abstract

摘要：

储层渗透率是储层产能的一个重要影响因素。针对常规测井渗透率模型在孔隙连通性差的低渗砂岩储层预测精度不高的问题,提出利用深度置信网络算法结合常规测井曲线预测储层渗透率的方法。该方法利用灰色关联法对测井曲线进行了关联度分析,依据相关度排序选取了特征敏感测井曲线,结合深度置信网络的有监督学习调优与对比散度算法进行数据挖掘,建立了渗透率的预测模型。该模型在以往BP神经网络的基础上改善了局部优化的问题,提高了网络模型的训练效率与预测精度。预测模型的平均相对误差为9.1 %,相比常规渗透率模型,降低了20 %左右。通过对实际资料的处理应用,结合误差分析,表明该方法能够有效地提高低渗透储层渗透率的预测精度。

关键词: 渗透率, 测井曲线, 灰色关联分析法, 深度置信网络, 预测

Abstract:

Reservoir permeability is an important factor affecting reservoir productivity. In order to solve the problem of poor prediction accuracy of conventional permeability logging models in low-permeability sandstone reservoirs with poor pore connectivity, a scheme combined with deep belief network（DBN） algorithm is proposed. First, the gray correlation method is used for the correlation analysis of logging curve, and according to the correlation ranking, the characteristic sensitive curves is sorted. Then, the optimization by supervised learning is combined with the contrastive divergence for the data mining to establish the prediction model of permeability. Compared to the previous BP neural network, DBN model improves the local optimization, and enhances the training efficiency and prediction accuracy. The average relative error of the prediction model is 9.1 %, which is about 20 % lower than that of the conventional permeability model. Based on the actual data processing applications and the error analysis, it is found that this method can effectively improve the prediction accuracy of permeability for the low permeability reservoirs.

Key words: permeability, logging curve, grey relational analysis, deep belief network, prediction

中图分类号:

TE31

赵军,张涛,何胜林,张桓荣,韩东,汤翟. 基于参数优选的储层渗透率深度置信网络模型预测初探[J]. 油气藏评价与开发, 2021, 11(4): 577-585.

ZHAO Jun,ZHANG Tao,HE Shenglin,ZHANG Huanrong,HAN Dong,TANG Di. Prediction of reservoir permeability by deep belief network based on optimized parameters[J]. Petroleum Reservoir Evaluation and Development, 2021, 11(4): 577-585.

图/表 16

图1

图2

表1

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图13

图14

表2

参考文献 24

[1]	赵久玉, 王付勇, 杨坤. 致密砂岩分形渗透率模型构建及关键分形参数计算方法[J]. 特种油气藏, 2020, 27(4):73-78.
	ZHAO Jiuyu, WANG Fuyong, YANG Kun. Tight sandstone fractal permeability model and key fractal parameter calculation[J]. Special Oil & Gas Reservoirs, 2020, 27(4):73-78.
[2]	程辉, 王付勇, 宰芸, 等. 基于高压压汞和核磁共振的致密砂岩渗透率预测[J]. 岩性油气藏, 2020, 32(3):122-132.
	CHENG Hui, WANG Fuyong, ZAI Yun, et al. Prediction of tight sandstone permeability based on high-pressure mercury intrusion(HPMI)and nuclear magnetic resonance(NMR)[J]. Lithologic Reservoirs, 2020, 32(3):122-132.
[3]	赵天逸, 宁正福, 陈刚, 等. 致密砂岩储集层渗透率预测修正方法[J]. 新疆石油地质, 2020, 41(3):337-343.
	ZHAO Tianyi, NING Zhengfu, CHEN Gang, et al. Modified methods of permeability prediction for tight sandstone reservoirs[J]. Xinjiang Petroleum Geology, 2020, 41(3):337-343.
[4]	张恒荣, 何胜林, 吴进波, 等. 一种基于Kozeny-Carmen方程改进的渗透率预测新方法[J]. 吉林大学学报(地球科学版), 2017, 47(3):899-906.
	ZHANG Hengrong, HE Shenglin, WU Jinbo, et al. A new method for predicting permeability based on modified Kozeny-Carmen[J]. Journal of Jilin University(Earth Science Edition), 2017, 47(3):899-906.
[5]	于华, 令狐松, 王谦, 等. 一种砂岩储层渗透率计算新方法[J]. 西南石油大学学报(自然科学版), 2020, 42(2):125-132.
	YU Hua, LINGHU Song, WANG Qian, et al. A new method for calculating permeability of sandstone reservoir[J]. Journal of Southwest Petroleum University(Science & Technology Edition), 2020, 42(2):125-132.
[6]	张冲. 基于海上砂砾岩低渗透率成因分析及测井评价[J]. 测井技术, 2019, 43(5):524-530.
	ZHANG Chong. Log evaluation of offshore low permeability conglomerate based on permeability genesis analysis[J]. Well Logging Technology, 2019, 43(5):524-530.
[7]	李奇, 高树生, 刘华勋, 等. 岩心渗透率的计算方法与适用范围[J]. 天然气工业, 2015, 35(3):68-73.
	LI Qi, GAO Shusheng, LIU Huaxun, et al. Core permeability calculation methods and application scopes[J]. Natural Gas Industry, 2015, 35(3) : 68-73.
[8]	毛志勇, 黄春娟, 路世昌, 等. 基于APSO-WLS-SVM的含瓦斯煤渗透率预测模型[J]. 煤田地质与勘探, 2019, 47(2):66-71.
	MAO Zhiyong, HUANG Chunjuan, LU Shichang, et al. Model of gas-bearing coal permeability prediction based on APSO-WLS-SVM[J]. Coal Geology & Exploration, 2019, 47(2):66-71.
[9]	邵良杉, 马寒. 煤体瓦斯渗透率的PSO-LSSVM预测模型[J]. 煤田地质与勘探, 2015, 43(4):23-26.
	SHAO Liangshan, MA Han. Model of coal gas permeability prediction based on PSO-LSSVM[J]. Coal Geology & Exploration, 2015, 43(4):23-26.
[10]	古勇. 基于改进支持向量机的煤体瓦斯渗透率预测[J]. 数学的实践与认识, 2016, 46(20):149-155.
	GU Yong. Prediction of coal gas permeability based on PSOSVM[J]. Mathematics in Practice and Theory, 2016, 46(20):149-155.
[11]	KABIRU O A, TAOREED O O, SUNDAY O O, et al. A hybrid particle swarm optimization and support vector regression model for modelling permeability prediction of hydrocarbon reservoir[J]. Journal of Petroleum Science and Engineering, 2017, 150(2):43-53. doi: 10.1016/j.petrol.2016.11.033
[12]	汪雷, 林亮, 李晶晶, 等. 基于测井信息的煤储层渗透率BP神经网络预测方法[J]. 煤炭科学技术, 2015, 43(7):122-126.
	WANG Lei, LIN Liang, LI Jingjing, et al. Method to predict permeability of coal reservoir with bp neural network based on logging information[J]. Coal Science and Technology, 2015, 43(7):122-126.
[13]	张言辉. 基于物性预测相对渗透率的改进神经网络方法[J]. 天然气与石油, 2020, 38(3):44-49.
	ZHANG Yanhui. Improved neural network method for predicting relative permeability based on physical properties[J]. Natural Gas and Oil, 2020, 38(3):44-49.
[14]	朱林奇, 张冲, 何小菊, 等. 基于改进BPNN与T₂全谱的致密砂岩储层渗透率预测[J]. 石油物探, 2017, 56(5):727-734.
	ZHU Linqi, ZHANG Chong, HE Xiaoju, et al. Permeability prediction of tight sandstone reservoir based on improved BPNN and T₂ full-spectrum[J]. Geophysical Prospecting for Petroleum, 2017, 56(5):727-734.
[15]	马晟翔, 李希建. 基于因子分析与BP神经网络的煤体瓦斯渗透率预测[J]. 煤矿开采, 2018, 23(6):108-111.
	MA Shengxiang, LI Xijian. Forecast of coal body gas permeability based on factor analysis and BP neural net[J]. Coal Mining Technology, 2018, 23(6):108-111.
[16]	BARAKA M N, CHUANNO S, SOLOMON A O, et al. Prediction of permeability using group method of data handling(GMDH) neural network from well log data[J]. Energies, 2020, 13(6):1-18. doi: 10.3390/en13010001
[17]	HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786):504-507. doi: 10.1126/science.1127647
[18]	王俊, 曹俊兴, 尤加春, 等. 基于门控循环单元神经网络的储层孔渗饱参数预测[J]. 石油物探, 2020, 59(4):616-627.
	WANG Jun, CAO Junxing, YOU Jiachun, et al. Prediction of reservoir porosity permeability and saturation based on a gated recurrent unit neural network[J]. Geophysical Prospecting for Petroleum, 2020, 59(4):616-627.
[19]	OLEG S, EVGENY B, DMITRY K. Driving digital rock towards machine learning: predicting permeability with gradient boosting and deep neural networks[J]. Computers and Geosciences, 2019, 127(6):91-98. doi: 10.1016/j.cageo.2019.02.002
[20]	JIAN W T, CHONG C Q, YING F S, et al. Surrogate permeability modelling of low permeable rocks using convolutional neural networks[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 366(13):103-113.
[21]	王小艺, 李柳生, 孔建磊, 等. 基于深度置信网络-多类模糊支持向量机的粮食供应链危害物风险预警[J]. 食品科学, 2020, 41(19):17-24.
	WANG Xiaoyi, LI Liusheng, KONG Jianlei, et al. Risk pre-warning of hazardous materials in cereal supply chain based on deep belief network-multiclass fuzzy support vector machine(DBN-MFSVM)[J]. Food Science, 2020, 41(19):17-24.
[22]	孟智慧. 基于深度置信网络的互联网流量预测方法[J]. 电信工程技术与标准化, 2020, 33(10):42-47.
	MENG Zhihui. Internet traffic prediction method based on deep belief network[J]. Telecom Engineering Technics and Standardization, 2020, 33(10):42-47.
[23]	许若冰, 王璇, 赵倩宇, 等. 基于卷积神经网络和深度置信网络的多类型能源需求预测方法[J]. 供用电, 2020, 37(10):65-70.
	XU Ruobing, WANG Xuan, ZHAO Qianyu, et al. A multi-energy demand prediction method based on convolutional neural network and deep belief network[J]. Distribution & Utilization, 2020, 37(10):65-70.
[24]	叶绍泽, 曹俊兴, 吴施楷, 等. 基于深度置信网络的总有机碳含量预测方法[J]. 地球物理学进展, 2018, 33(6):2490-2497.
	YE Shaoze, CAO Junxing, WU Shikai, et al. Prediction method of total organic carbon content based on deep belief nets[J]. Progress in Geophysics, 2018, 33(6):2490-2497.

测井类别	关联度	排序
声波时差（AC）	0.874 7	2
密度（DEN）	0.892 5	1
自然伽马（GR）	0.723 0	3
补偿中子（CNL）	0.667 9	4
井径（CAL）	0.507 7	8
自然电位（SP）	0.557 7	7
浅侧向（RILM）	0.656 4	5
深侧向（RILD）	0.648 6	6

井号	深度段（m）	常规孔隙度预测模型（%）	BP神经网络模型（%）	深度置信网络模型（%）
WC1井	3 663.3~3 678.9	32.70	22.30	10.60
WC2井	3 849.7~3 867.7	28.40	18.40	7.20
WC3井	3 987.8~4 006.1	30.30	19.20	8.14
WC4井	3 851~3 855.4	28.10	19.90	9.25
WC5井	3 753.6~3 771.6	31.20	20.10	10.40
平均值		30.14	19.98	9.12