基于改进SSA-BPNN的煤层气直井井底流压预测研究

doi:10.13809/j.cnki.cn32-1825/te.2025.02.009

摘要/Abstract

摘要：

煤层气资源广泛应用直井开发，采用控压控水的排采制度，井底流压是排采方案设计与设备选型的重要参数，因此，煤层气直井井底流压预测具有重要的意义。为了便捷、准确地预测煤层气直井井底流压，指导煤层气井的控压排采，引入机器学习领域中的反向传播神经网络（BPNN）模型，同时对麻雀搜索算法（SSA）进行改进，耦合构建基于改进麻雀搜索算法-反向传播神经网络（SSA-BPNN）的煤层气直井井底流压预测模型。选取了生产现场常规测量的5个影响井底流压的参数作为井底流压预测模型的输入参数，相对应的井底流压数值作为井底流压预测模型的输出参数。将600组实测数据划分为训练集、验证集与测试集，完成了煤层气直井井底流压预测模型的建立与校验工作。BPNN模型与改进SSA-BPNN模型的验证集平均绝对百分比误差分别为3.10%与0.53%，可以看出利用改进SSA与BPNN的耦合建模，能够解决BPNN易陷于局部最优的问题，提高了煤层气直井井底流压的预测精度。同时将改进SSA-BPNN模型与遗传算法-支持向量回归机（GA-SVR）模型和物理模型解析方法进行对比，结果显示：3种不同模型的平均绝对百分比误差分别为1.318%、4.971%、18.156%，改进SSA-BPNN模型的误差最低，且在井底流压较低时，改进SSA-BPNN模型的预测精度显著提高，展现出较高的准确性与良好的适用性。改进SSA-BPNN模型仅需5个输入参数，减少了输入与计算参数的复杂度，且无须考虑井筒内流体分布情况，可覆盖排采各阶段，在不同压力区间都有较高准确性。

关键词: 煤层气, 麻雀搜索算法, 神经网络, 井底流压, 预测模型

Abstract:

Coalbed methane resources are extensively developed using vertical wells, with controlled-pressure and controlled-water drainage systems. The flowing bottom hole pressure is a crucial parameter in the design of drainage schemes and equipment selection. Therefore, it is of great significance to predict the flowing bottom hole pressure for vertical coalbed methane wells. To conveniently and accurately forecast the flowing bottom hole pressure of vertical coalbed methane and guide their pressure control and drainage, a Backpropagation Neural Network (BPNN) algorithm from the field of machine learning was introduced. Additionally, the Sparrow Search Algorithm (SSA) was improved. These were coupled to establish a forecasting model for flowing bottom hole pressure based on the improved SSA-BPNN approach. Five routinely measured parameters that influence flowing bottom hole pressure were selected as the input parameters for the prediction model, with corresponding bottom hole pressure values as the output parameters. A total of 600 sets of field-measured data were partitioned into training, validation, and testing datasets to develop and validate the forecasting model for vertical coalbed methane wells. The validation set showed that the mean absolute percentage errors for the BPNN model and the Improved SSA-BPNN model on the validation set were 3.10% and 0.53%, respectively. This demonstrated that coupling the Improved SSA and BPNN effectively overcame the propensity of BPNN to converge to local optima, thereby improving the prediction accuracy of flowing bottom hole pressure in a vertical coalbed methane well. Furthermore, the improved SSA-BPNN model was compared with the Genetic Algorithm-Support Vector Regression (GA-SVR) method and the physical model-based analytical method. The results revealed that the mean absolute percentage errors for these three different models were 1.318%, 4.971%, and 18.156%, respectively. The Improved SSA-BPNN model had the lowest error, and its prediction accuracy significantly improved when the flowing bottom hole pressure was low, demonstrating its higher accuracy and strong applicability. The Improved SSA-BPNN model requires only five input parameters, reducing the complexity of input and calculation parameters. It does not require consideration of the fluid distribution within the wellbore and can cover all stages of drainage, maintaining high accuracy across different pressure ranges.

Key words: coalbed methane, sparrow search algorithm, neural network, bottom hole flowing pressure, prediction model

中图分类号:

TE37

余洋,董银涛,李云波等. 基于改进SSA-BPNN的煤层气直井井底流压预测研究[J]. 油气藏评价与开发, 2025, 15(2): 250-256.

YU Yang,DONG Yintao,LI Yunbo, et al. Research on prediction of bottom hole flowing pressure for vertical coalbed methane wells based on improved SSA-BPNN[J]. Petroleum Reservoir Evaluation and Development, 2025, 15(2): 250-256.

图/表 7

图1

图2

表1

图3

图4

表2

图5

参考文献 30

[1]	邹才能, 杨智, 黄士鹏, 等. 煤系天然气的资源类型、形成分布与发展前景[J]. 石油勘探与开发, 2019, 46(3): 433-442.
	ZOU Caineng, YANG Zhi, HUANG Shipeng, et al. Resource types, formation, distribution and prospects of coal-measure gas[J]. Petroleum Exploration & Development, 2019, 46(3): 433-442.
[2]	娄钰, 潘继平, 王陆新, 等. 中国天然气资源勘探开发现状、问题及对策建议[J]. 国际石油经济, 2018, 26(6): 21-27.
	LOU Yu, PAN Jiping, WANG Luxin, et al. Problems and countermeasures in the exploration and development of natural gas resources in China[J]. International Petroleum Economics, 2018, 26(6): 21-27.
[3]	赵贤正, 朱庆忠, 孙粉锦, 等. 沁水盆地高阶煤层气勘探开发实践与思考[J]. 煤炭学报, 2015, 40(9): 2131-2136.
	ZHAO Xianzheng, ZHU Qingzhong, SUN Fenjin, et al. Practice of coalbed methane exploration and development in Qinshui Basin[J]. Journal of China Coal Society, 2015, 40(9): 2131-2136.
[4]	张遂安, 曹立虎, 杜彩霞. 煤层气井产气机理及排采控压控粉研究[J]. 煤炭学报, 2014, 39(9): 1927-1931.
	ZHANG Suian, CAO Lihu, DU Caixia, et al. Study on CBM production mechanism and control theory of bottom-hole pressure and coal fines during CBM well production[J]. Journal of China Coal Society, 2014, 39(9): 1927-1931.
[5]	刘斌, 杜海为, 崔金榜, 等. 煤层气井排采控制技术发展现状与展望[J]. 石油钻采工艺, 2019, 41(4): 489-493.
	LIU Bin, DU Haiwei, CUI Jinbang, et al. Development status and prospect of CBM well production control technologies[J]. Oil Drilling & Production Technology, 2019, 41(4): 489-493.
[6]	杜新锋, 郭盛强, 张群, 等. 多煤层煤层气井分层控压合层排采技术及装备[J]. 煤炭科学技术, 2018, 46(6): 114-118.
	DU Xinfeng, GUO Shengqiang, ZHANG Qun, et al. Separate－layer pressure control and multi－layer drainage technology and device for coalbed methane wells with multiple seams[J]. Coal Science and Technology, 2018, 46(6): 114-118.
[7]	刘新福, 綦耀光, 李延祥, 等. 单相流煤层气井井底流压预测方法研究[J]. 中国矿业大学学报, 2010, 39(6): 876-880.
	LIU Xinfu, QI Yaoguang, LI Yanxiang, et al. Prediction of flowing bottomhole pressures for single-phase coalbed methane wells[J]. Journal of China University of Mining & Technology, 2010, 39(6): 876-880.
[8]	孙仁远, 宣英龙, 任晓霞, 等. 煤层气井井底流压计算方法[J]. 石油钻采工艺, 2012, 34(4): 100-103.
	SUN Renyuan, XUAN Yinglong, REN Xiaoxia, et al. Methods of bottom-hole flow pressure calculations for coalbed methane wells[J]. Oil Drilling & Production Technology, 2012, 34(4): 100-103.
[9]	杨焦生, 王一兵, 王宪花. 煤层气井井底流压分析及计算[J]. 天然气工业, 2010, 30(2): 66-68.
	YANG Jiaosheng, WANG Yibing, WANG Xianhua. Analysis and computation of flowing bottom hole pressure in coalbed methane wells[J]. Natural Gas Industry, 2010, 30(2): 66-68.
[10]	张永平, 孟召平, 刘贺, 等. 煤层气井排采初期井底流压动态模型及应用分析[J]. 煤田地质与勘探, 2016, 44(2): 29-33.
	ZHANG Yongping, MENG Zhaoping, LIU He, et al. Dynamic model for bottomhole flowing pressure in initial stage of CBM wells drainage and its application[J]. Coal Geology & Exploration, 2016, 44(2): 29-33.
[11]	刘新福, 綦耀光, 刘春花, 等. 气水两相煤层气井井底流压预测方法[J]. 石油学报, 2010, 31(6): 998-1003.
	LIU Xinfu, QI Yaoguang, LIU Chunhua, et al. Prediction of flowing bottomhole pressures for two-phase coalbed methane wells[J]. Acta Petrolei Sinica, 2010, 31(6): 998-1003.
[12]	赵金, 张遂安. 煤层气井底流压生产动态研究[J]. 煤田地质与勘探, 2013, 41(2): 21-24.
	ZHAO Jin, ZHANG Suian. Production dynamics of CBM bottom hole pressure[J]. Coal Geology & Exploration, 2013, 41(2): 21-24.
[13]	毛慧, 韩国庆, 吴晓东, 等. 确定煤层气井合理生产压差的新思路[J]. 天然气工业, 2011, 31(3): 52-55.
	MAO Hui, HAN Guoqing, WU Xiaodong, et al. A new discussion on the determination methods of reasonable drawdown pressure for a coalbed methane gas well[J]. Natural Gas Industry, 2011, 31(3): 52-55.
[14]	董银涛, 鞠斌山, 张遂安. 煤层气直井井筒环空压力模型[J]. 煤炭学报, 2018, 43(9): 2534-2542.
	DONG Yintao, JU Binshan, ZHANG Suian. Wellbore annulus pressure model of the vertical coalbed methane well[J]. Journal of China Coal Society, 2018, 43(9): 2534-2542.
[15]	黄家宸, 张金川. 机器学习预测油气产量现状[J]. 油气藏评价与开发, 2021, 11(4): 613-620.
	HUANG Jiachen, ZHANG Jinchuan. Overview of oil and gas production forecasting by machine learning[J]. Petroleum Reservoir Evaluation and Development, 2021, 11(4): 613-620.
[16]	韩力群, 施彦. 人工神经网络理论、设计及应用[M]. 北京: 化学工业出版社, 2002.
	HAN Liqun, SHI Yan. Artificial neural network theory, design and application[M]. Beijing: Chemical Industry Press, 2002.
[17]	沙林秀, 胥陈卓. 基于主成分分析的NCPSO-BP机械钻速预测[J]. 石油钻采工艺, 2022, 44(4): 515-521.
	SHA Linxiu, XU Chenzhuo. Prediction of NCPSO-BP ROP based on principal component ana[J]. Oil Drilling & Production Technology, 2022, 44(4): 515-521.
[18]	王虹, 尤秀松, 李首滨, 等. 基于遗传算法与BP神经网络的支架跟机自动化研究[J]. 煤炭科学技术, 2021, 49(1): 272-277.
	WANG Hong, YOU Xiusong, LI Shoubin, et al. Research on automation of support based on genetic algorithm and BP neural network[J]. Coal Science and Technology, 2021, 49(1): 272-277.
[19]	张金梦, 刘慧君. 遗传算法优化BP神经网络的泊车位数量预测[J]. 重庆大学学报, 2018, 41(3): 80-85.
	ZHANG Jinmeng, LIU Huijun. Prediction of spare parking spaces based on BP neural network optimized by genetic algorithm[J]. Journal of Chongqing University, 2018, 41(3): 80-85.
[20]	臧子婧, 吴海波, 张平松, 等. 基于ABC-BP模型的煤层含气量预测[J]. 煤田地质与勘探, 2021, 49(2): 152-158.
	ZANG Zijing, WU Haibo, ZHANG Pingsong, et al. Prediction of coal seam gas content based on ABC-BP model[J]. Coal Geology & Exploration, 2021, 49(2): 152-158.
[21]	XUE J K, SHEN B. A novel swarm intelligence optimization approach: sparrow search algorithm[J]. Systems Science & Control Engineering An Open Access Journal, 2020, 8(1): 22-34.
[22]	赵涛, 于师建. 基于GA-BP神经网络算法的高密度电法非线性反演[J]. 煤田地质与勘探, 2017, 45(2): 147-151.
	ZHAO Tao, YU Shijian. GA-BP neural network algorithm-based nonlinear inversion for high density resistivity method[J]. Coal Geology & Exploration, 2017, 45(2): 147-151.
[23]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444.
[24]	罗发强, 刘景涛, 陈修平, 等. 基于BP和LSTM神经网络的顺北油田5号断裂带地层孔隙压力智能预测方法[J]. 石油钻采工艺, 2022, 44(4): 506-514.
	LUO Faqiang, LIU Jingtao, CHEN Xiuping. Intelligent method for predicting formation pore pressure in No. 5 fault zone in Shunbei oilfield based on BP and LSTM neural network [J]. Oil Drilling & Production Technology, 2022, 44(4): 506-514.
[25]	刘建军, 石定元, 武国宁. 基于Kent映射的混合混沌优化算法[J]. 计算机工程与设计, 2015, 36(6): 1498-1503.
	LIU Jianjun, SHI Dingyuan, WU Guoning. Hybrid chaotic optimization algorithm based on Kent map[J]. Computer Engineering And Design, 2015, 36(6): 1498-1503.
[26]	周刘喜, 张兴华, 李纬. 一种改进的多目标粒子群优化算法[J]. 计算机工程与应用, 2009, 45(33): 38-41.
	ZHOU Liuxi, ZHANG Xinghua, LI Wei. Improved multi-objective particle swarm optimization algorithm[J]. Computer Engineering and Applications, 2009, 45(33): 38-41.
[27]	李永恒, 赵志刚. 基于越界重置和高斯变异的蝙蝠优化算法[J]. 计算机工程与科学, 2019, 41(1): 144-152.
	LI Yongheng, ZHAO Zhigang. An improved bat algorithm based on cross-border relocation and Gaussian mutation[J]. Computer Engineering & Science, 2019, 41(1): 144-152.
[28]	吕鑫, 慕晓冬, 张钧, 等. 混沌麻雀搜索优化算法[J]. 北京航空航天大学学报, 2021, 47(8): 1712-1720.
	LV Xin, MU Xiaodong, ZHANG Jun, et al. Chaos sparrow search optimization algorithm[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(8): 1712-1720.
[29]	LUO Z, HASANIPANAH M, AMNIEH H B. GA-SVR: a novel hybrid data-driven model to simulate vertical load capacity of driven piles[J]. Engineering with Computers, 2021, 37(12): 823-831.
[30]	孙雷, 罗强, 潘毅, 等. 基于GA-SVR的CO₂驱原油最小混相压力预测模型[J]. 大庆石油地质与开发, 2017, 36(3): 123-129.
	SUN Lei, LUO Qiang, PAN Yi, et al. Predicting model of the oil minimal miscible pressure for the CO₂ flooding based on GA-SVR[J]. Petroleum Geology & Oilfield Development in Daqing, 2017, 36(3): 123-129.

井口套压/MPa	动液面高度/m	动液面深度/m	日产气量/m³	日产水量/m³	井底流压/MPa
0.09	45.29	563.37	12 772.69	24.21	0.54
0.08	65.08	543.58	7 452.36	4.20	0.71
0.10	52.63	556.03	10 896.58	8.50	0.62
0.15	16.72	410.45	4 151.44	18.00	0.31
0.09	62.99	364.18	2 026.69	10.83	0.71
0.13	9.12	424.43	235.92	5.37	0.21
0.20	66.94	364.22	260.09	1.89	0.86
0.14	302.00	361.10	0	4.00	3.18
1.25	17.90	845.00	501.00	1.30	1.40
0.05	64.47	544.19	9 152.99	23.86	0.68
0.15	11.67	421.88	232.83	5.34	0.26
1.39	11.10	300.00	3 650.00	0.90	1.39
2.62	3.20	659.90	158.00	3.90	2.73
0.08	185.40	600.30	0	7.40	1.92

预测模型	均方误差	平均绝对误差	平均绝对百分比误差/%
改进SSA-BPNN模型	2.52×10^-4	0.013	1.318
GA-SVR模型	5.54×10^-4	0.027	4.971
文献[14]中解析模型	1.55×10^-3	0.087	18.156