基于神经网络的南海东部砂岩油藏采收率预测方法

doi:10.13809/j.cnki.cn32-1825/te.2021.05.010

摘要/Abstract

摘要：

目前南海东部砂岩油藏采收率预测多采用数值模拟和线性回归等方法,这些方法分别存在耗时长和精度低的缺点。为了快速、准确地预测油藏采收率,选择50个已开发油藏作为数据样本,在利用主成分分析对采收率影响因素进行特征提取的基础上,运用神经网络回归法,建立了适用于南海东部海相砂岩油藏的采收率预测模型。通过与支持向量机回归和线性回归两种方法建立的采收率预测模型的预测结果对比表明,神经网络回归模型预测结果具有较高的预测精度,能够快速评价此类油藏的开发潜力。

关键词: 神经网络, 南海东部, 砂岩油藏, 采收率, 预测模型, 主成分分析

Abstract:

At present, the methods such as numerical simulation and linear regression are mostly used to predict the oil recovery of sandstone reservoirs in the eastern South China Sea, but some of them are time-consuming or with low accuracy. In order to predict the oil recovery quickly and accurately, 50 developed reservoirs are selected as the data samples. Based on the feature extraction of the influencing factors of the principal component analysis on recovery, a recovery prediction model suitable for sandstone reservoirs in the eastern South China Sea is established by the neural network. Compared with that of two methods of support vector machine regression and linear regression, the prediction results of neural network regression model have high prediction accuracy, which can evaluate the development potential of the similar reservoirs quickly.

Key words: neural network, eastern South China Sea, sandstone reservoir, oil recovery, prediction model, principal component analysis

中图分类号:

TE327

李伟,唐放,侯博恒等. 基于神经网络的南海东部砂岩油藏采收率预测方法[J]. 油气藏评价与开发, 2021, 11(5): 730-735.

Wei LI,Fang TANG,Boheng HOU, et al. A method for oil recovery prediction of sandstone reservoirs in the eastern South China Sea based on neural network[J]. Petroleum Reservoir Evaluation and Development, 2021, 11(5): 730-735.

图/表 10

图1

表1

神经网络回归训练样本数据"

油藏名称	含油面积（km²）	油层厚度（m）	孔隙度（%）	含油饱和度（%）	地层压力（MPa）	流度（10^-3 μm²/mPa·s）	井网密度（口/km²）	渗透率（10^-3 μm²）	地质储量（10⁴ m³）	采收率（%）
N1	7.2	3.5	15.8	61.0	28.6	822.1	0.98	271.3	164	76.5
N2	7.5	6.4	16.2	63.4	29.0	737.9	0.79	317.3	368	67.8
N3	8.9	2.0	16.5	61.1	29.0	656.5	0.13	282.3	134	64.5
N4	7.8	2.9	22.2	63.6	18.5	196.2	0.14	1 436.4	305	47.9
N5	5.5	5.2	15.4	61.6	29.2	534.2	1.09	192.3	193	73.4
N6	9.8	5.6	21.6	60.6	14.3	138.5	1.96	637.0	681	61.9
N7	6.4	2.8	19.9	46.5	16.0	143.5	3.21	660.0	149	62.2
N8	3.7	3.4	20.5	53.2	16.4	123.7	2.70	569.0	127	51.9
N9	10.9	5.6	21.3	75.8	20.1	345.4	2.11	1 088.0	912	67.0
$?$	$?$	$?$	$?$	$?$	$?$	$?$	$?$	$?$	$?$	$?$
N50	2.6	3.3	23.0	63.2	20.2	651.0	0.50	2 929.7	129	44.4

表1

图2

表2

表3

表4

表5

表6

图3

图4

参考文献 21

[1]	钱其豪. 海上砂岩油藏高速开发驱替特征[D]. 北京:中国石油大学(北京),2011.
	QIAN Qihao. Marine sandstone reservoir high-speed development law[D]. Beijing: China University of Petroleum (Beijing), 2011.
[2]	王锋. 盘油田复杂断块油藏注采井网优化[J]. 西安石油大学学报: 自然科学版, 2014, 29(5):58-60.
	WANG Feng. Optimization of reasonable injection-production well pattern in complex fault block of Linpan Oilfield[J]. Journal of Xi’an Shiyou University: Natural Science Edition, 2014, 29(5):58-60.
[3]	任耀宇, 张弦, 罗鹏飞, 等. 特低渗透轻质油藏热水驱提高采收率试验研究[J]. 非常规油气, 2019, 6(1):69-74.
	REN Yaoyu, ZHANG Xian, LUO Pengfei, et al. An experimental study of hot water flood for enhancing oil recovery in ultra-low permeability reservoir[J]. Unconventional Oil & Gas, 2019, 6(1):69-74.
[4]	贾凯锋, 计董超, 高金栋, 等. 低渗透油藏CO₂驱油提高原油采收率研究现状[J]. 非常规油气, 2019, 6(1):107-114.
	JIA Kaifeng, JI Dongchao, GAO Jindong, et al. The exisiting state of enhanced oil recovery by CO₂ flooding in low permeability reservoirs[J]. Unconventional Oil & Gas, 2019, 6(1):107-114.
[5]	冯沙沙, 王亚会, 李伟, 等. 南海东部砂岩油藏采收率经验公式的确定[J]. 石油化工应用, 2015, 34(1):74-77.
	FENG Shasha, WANG Yahui, LI Wei, et al. The determination of recovery factor empirical formula in Nanhai east sandstone reservoirs[J]. Petrochemical Industry Application, 2015, 34(1):74-77.
[6]	段宇, 杨东东, 王美楠, 等. 渤海水驱砂岩油藏采收率经验公式研究[J]. 石化技术, 2018, 25(9):146-147.
	DUAN Yu, YANG Dongdong, WANG Meinan, et al. The research of the method on recovery efficiency empirical formula of water-drive sandstone reservoir of Bohai oil field[J]. Petrochemical Industry Technology, 2018, 25(9):146-147.
[7]	鲁瑞彬, 刘双琪, 胡琳, 等. 水驱砂岩油田动态采收率计算新方法[J]. 西安石油大学学报(自然科学版), 2019, 34(4):60-66.
	LU Ruibin, LIU Shuangqi, HU Lin, et al. A new method for calculating dynamic recovery factor of water-flooding sandstone oilfield[J]. Journal of Xi’an Shiyou University (Natural Science Edition), 2019, 34(4):60-66.
[8]	吴春新. 基于RS-LSSVM水驱井组采出程度计算模型——以渤海黄河口凹陷为例[J]. 新疆石油天然气, 2019, 15(3):49-53.
	WU Chunxin. Calculation model of water flooding well group recovery degree based on RS-LSSVM——Taking Huanghe River sag in Bohai as example[J]. Xinjiang Oil & Gas, 2019, 15(3):49-53.
[9]	王敏, 陈民锋, 刘广为, 等. 主成分分析法确定海上油田水驱效果评价关键指标[J]. 油气地质与采收率, 2015, 22(2):112-116.
	WANG Min, CHEN Minfeng, LIU Guangwei, et al. Application of principal component analysis on determining the key evaluation indicators of water flooding effects in offshore oilfield[J]. Petroleum Geology and Recovery Efficiency, 2015, 22(2):112-116.
[10]	许晓明, 李彦兰, 孙景民. 基于模糊数学的油藏干层识别研究[J]. 西南石油大学学报(自然科学版), 2019, 41(2):45-52.
	XU Xiaoming, LI Yanlan, SUN Jingmin. Study on identification of dry layers based on fuzzy mathematics[J]. Journal of Southwest Petroleum University(Science & Technology Edition), 2019, 41(2):45-52.
[11]	WANG J K, HU J H, ZHANG Y, et al. Investigation of imbibition areas during well shut-in based on mercury injection experiment and BP neural network[J]. Fuel, 2019, 254: 115621.1-115621.8.
[12]	唐振国, 迟博, 吕金龙, 等. 基于多属性神经网络地震相的扶余油层砂体精细刻画及应用[J]. 非常规油气, 2020, 7(2):41-48.
	TANG Zhenguo, CHI Bo, LYU Jinlong, et al. Fine characterization and application of sand body in Fuyu reservoir based on seismic facies by multi-attribute neural network[J]. Unconventional Oil & Gas, 2020, 7(2):41-48.
[13]	吴君达, 李治平, 孙妍, 等. 基于神经网络的剩余油分布预测及注采参数优化[J]. 油气地质与采收率, 2020, 27(4):85-93.
	WU Junda, LI Zhiping, SUN Yan, et al. Neural network-based prediction of remaining oil distribution and optimization of injection-production parameters[J]. Petroleum Geology and Recovery Efficiency, 2020, 27(4):85-93.
[14]	杨丽娜, 解国军. 油气资源丰度预测的人工神经网络方法——以济阳坳陷为例[J]. 石油天然气学报, 2007, 29(1):55-58.
	YANG Lina, XIE Guojun. Prediction of abundance of hydrocarbon resources with artificial neural network method——Taking Jiyang depression for example[J]. Journal of Oil and Gas Technology, 2007, 29(1):55-58.
[15]	刘苏苏, 孙立民. 支持向量机与RBF神经网络回归性能比较研究[J]. 计算机工程与设计, 2011, 32(12):4202-4205.
	LIU Susu, SUN Limin. Performance comparison of regression prediction on support vector machine and RBF neural network[J]. Computer Engineering and Design, 2011, 32(12):4202-4205.
[16]	刘文超, 卢祥国, 刘进祥, 等. 一种基于BP神经网络的调驱增油预测方法[J]. 西安石油大学学报(自然科学版), 2012, 27(1):47-52.
	LIU Wenchao, LU Xiangguo, LIU Jinxiang, et al. Prediction method of oil increment of profile-control and flooding measures using BP neural network based on core flooding parameters[J]. Journal of Xi’an Shiyou University(Natural Science Edition), 2012, 27(1):47-52.
[17]	ZHOU Deqiang. Grey verhulst model based on BP neural network optimization for oil production forecasting[J]. International Journal of Energy Science, 2012, 2(3):115-118.
[18]	ZHOU Yingming, WANG Qiushi, WANG Peng, et al. Moisture content of crude oil based on BP neural network algorithm to predict in oilfield[J]. Applied Mechanics & Materials, 2012, 170-173:1290-1293.
[19]	YU S W, ZHU K J, DIAO F Q. A dynamic all parameters adaptive BP neural networks model and its application on oil reservoir prediction[J]. Applied Mathematics and Computation, 2007, 195(1):66-75. doi: 10.1016/j.amc.2007.04.088
[20]	张烈平, 周德俭, 牛秦洲. 基于BP神经网络的预测建模系统的研究与实现[J]. 计算机仿真, 2004(9):48-50.
	ZHANG Lieping, ZHOU Dejian, NIU Qinzhou. Research and realization of forecasting model system based on BP neural network[J]. Computer Simulation, 2004(9):48-50.
[21]	王涛. 人工神经网络在CO₂驱采收率预测中的应用[J]. 特种油气藏, 2011, 18(4):77-79.
	WANG Tao. Application of artificial neural network in recovery factor forecast of carbon dioxide flooding[J]. Special Oil & Gas Reservoirs, 2011, 18(4):77-79.

特征编号	特征值	方差解释率（%）	累积方差解释率（%）
1	2.91	34.19	34.19
2	2.03	23.84	58.03
3	1.51	17.71	75.74
4	0.90	10.54	86.28
5	0.45	5.23	91.51
6	0.36	4.28	95.79
7	0.21	2.51	98.30
8	0.10	1.17	99.47
9	0.04	0.53	100.00

油藏名称	含油面积（km²）	油层厚度（m）	孔隙度（%）	含油饱和度（%）	地层压力（MPa）	流度（10^-3 μm²/mPa·s）	井网密度（口/km²）	渗透率（10^-3 μm²）	地质储量（10⁴ m³）	采收率（%）
H1	4.9	3.4	18.3	53.1	17.1	75.5	2.66	2 393.7	155	46.97
H2	5.2	4.4	24.2	75.6	18.2	369.9	4.42	3 587.7	402	53.87
H3	8.2	4.1	23.7	73.3	20.2	435.5	3.02	2 526.1	559	72.30
H4	9.1	4.3	25.0	71.8	22.3	534.2	3.13	3 098.3	668	69.72
H5	10.4	10.5	21.6	66.1	24.2	439.9	5.98	1 659.8	1 485	69.85
H6	2.4	2.0	21.9	58.8	19.6	284.3	1.60	1 648.7	60	64.56

油藏名称	实际采收率（%）	神经网络预测值（%）	相对误差（%）	支持向量机回归预测值（%）	相对误差（%）	线性回归预测值（%）	相对误差（%）
H1	46.97	47.23	0.55	51.77	10.22	49.60	5.60
H2	53.87	58.43	8.46	56.97	5.75	60.72	12.72
H3	72.30	63.71	11.88	56.10	22.41	60.00	17.01
H4	69.72	71.46	2.50	49.99	28.30	64.38	7.66
H5	69.85	73.55	5.30	64.42	7.77	77.84	11.44
H6	64.56	59.33	8.10	58.23	9.80	55.22	14.47

水平	含油面积（km²）	油层厚度（m）	孔隙度（%）	含油饱和度（%）	地层压力（MPa）	流度（10^-3 μm²/mPa·s）	井网密度（口/km²）	渗透率（10^-3 μm²）
1	2	2	15	0.60	16	60	0.5	500
2	5	5	18	0.65	20	300	2.5	1 500
3	8	8	21	0.70	24	540	4.5	2 500
4	11	11	24	0.75	28	780	6.5	3 500

分析项	偏差平方和	自由度	F值	F临界值	显著性	分析项	偏差平方和	自由度	F值	F临界值	显著性
含油面积	124.00	3.00	2.03	9.28	不显著	地层压力	61.00	3.00	1.00	9.28	不显著
油层厚度	3 324.00	3.00	54.60	9.28	显著	流度	852.00	3.00	14.00	9.28	显著
孔隙度	174.00	3.00	2.86	9.28	不显著	井网密度	1 489.00	3.00	24.47	9.28	显著
含油饱和度	131.00	3.00	2.15	9.28	不显著	渗透率	631.00	3.00	10.36	9.28	显著