刘向斌,钱坤,张团,孙延安,卢成国,郑东志
LIU Xiangbin,QIAN Kun,ZHANG Tuan,SUN Yan'an,LU Chengguo,ZHENG Dongzhi

• 1:多资源协同陆相页岩油绿色开采全国重点实验
• 2:中国石油大庆油田有限责任公司采油工艺研究院
• 3:黑龙江省油气藏增产增注重点实验室

摘要(Abstract)：

为攻克页岩油井产能预测精度不足及排采制度优化复杂的难题，通过融合多源数据特征分析，对异常数据进行归一化处理后构建样本集；随后基于随机森林与沙普利值（S_(SHAP)）分析识别关键驱动特征，并以气液比、油嘴直径、井底温度为主要输入参数构建堆叠集成（Stacking）预测框架；采用交叉验证生成元特征并由线性回归实现融合，进一步结合分阶段动态优化策略，对油嘴尺寸及气液比进行联合调控，从而实现排采制度的自适应优化。结果表明：Stacking集成方法在测试集上表现最优，决定系数（R~2）为0.927 0，平均绝对误差（B_(MAE)）为0.266 5，明显优于各单一模型，邻井验证显示其泛化能力稳定可靠；经动态调整优化后，见油放产和衰减阶段的累计产油量分别提升17.88%和11.16%。研究成果可为页岩油井排采制度优化策略制定提供有力支撑，有效提升产能预测精度与排采效率。
To address the challenges of insufficient accuracy in shale oil well productivity prediction and the complexity of production-drainage system optimization, a normalized sample dataset is constructed after processing abnormal data by integrating multi-sources data characteristics analysis. Key driving features are then identified through random forest and Shapley value(S_(SHAP)) analysis, and a Stacking ensemble prediction framework is established with gas-liquid ratio, choke diameter and bottom hole temperature as primary input parameters. Meta-features are generated via cross-validation and integrated through linear regression. A staged dynamic optimization strategy is subsequently applied to jointly regulate choke diameter and gas-liquid ratio, achieving adaptive production-drainage optimization. The results show that Stacking ensemble method achieves optimal performance on test set, with a determination coefficient(R~2) of 0.927 0 and mean absolute error(B_(MAE)) of 0.266 5, significantly outperforming individual models. Validation of adjacent wells demonstrates stable and reliable generalization capability. After dynamic adjustment optimization, cumulative oil production during oil breakthrough and decline stages increased by 17.87% and 11.16%, respectively. The research provides support for formulating shale oil well production-drainage optimization strategies, effectively improving productivity prediction accuracy and drainage efficiency.

关键词(KeyWords)： 页岩油;产能预测;机器学习;Stacking模型;排采优化
shale oil;productivity prediction;machine learning;Stacking model;production-drainage optimization

Abstract:

Keywords:

基金项目(Foundation): 中国石油天然气集团有限公司科技项目“陆相页岩油规模增储上产与勘探开发技术研究”（2023ZZ15YJ04）

作者(Author): 刘向斌,钱坤,张团,孙延安,卢成国,郑东志
LIU Xiangbin,QIAN Kun,ZHANG Tuan,SUN Yan'an,LU Chengguo,ZHENG Dongzhi

DOI: 10.19597/J.ISSN.1000-3754.202509055

参考文献(References)：

• [1]马先林，周德胜，蔡文斌，等.基于可解释机器学习的水平井产能预测方法[J].西南石油大学学报（自然科学版），2022,44(4):81-90.MA Xianlin,ZHOU Desheng,CAI Wenbin,et al. An interpretable machine learning approach to prediction horizontal well productivity[J]. Journal of Southwest Petroleum University(Science&Technology Edition),2022,44(4):81-90.
• [2] WANG S,GE W,ZHANG Y L,et al. Predicting the productivity of fractured horizontal wells using few-shot learning[J].Petroleum Science,2025,22(2):787-804.
• [3]蔡晖，石洪福，王昭，等.基于机器学习的疏松砂岩油藏定向井初期产能预测与影响因素分析[J].中国海上油气，2025,37(3):123-131.CAI Hui,SHI Hongfu,WANG Zhao,et al. Initial productivity prediction and influencing factor analysis of directional wells in unconsolidated sandstone reservoirs based on machine learning[J]. China Offshore Oil and Gas,2025,37(3):123-131.
• [4]罗水亮，漆影强，唐松，等.基于改进Stacking算法的碳酸盐岩储层测井岩性识别方法与应用[J].特种油气藏，2025,32(4):58-67.LUO Shuiliang, QI Yingqiang, TANG Song, et al. Lithology logging identification method and application to carbonate reservoirs based on improved Stacking algorithm[J].Special Oil&Gas Reservoirs,2025,32(4):58-67.
• [5]费世祥，崔越华，李小锋，等.鄂尔多斯盆地中、东部深层煤岩气水平井高效开发主控因素[J].石油与天然气地质，2025,46(1):273-287.FEI Shixiang, CUI Yuehua, LI Xiaofeng, et al. Main factors controlling the efficient production of horizontal wells for deep coal-rock gas in the eastern and central Ordos Basin[J].Oil&Gas Geology,2025,46(1):273-287.
• [6]史长林，魏莉，张剑，等.基于机器学习的储层预测方法[J].油气地质与采收率，2022,29(1):90-97.SHI Changlin,WEI Li,ZHANG Jian,et al. Reservoir prediction method based on machine learning[J].Petroleum Geology and Recovery Efficiency,2022,29(1):90-97.
• [7]秦志军，操应长，冯程.基于改进型随机森林算法的页岩岩性识别：以准噶尔盆地芦草沟组为例[J].新疆石油地质，2024,45(5):595-603.QIN Zhijun, CAO Yingchang, FENG Cheng. Shale lithology identification based on improved random forest algorithm:A case of Lucaogou Formation in Junggar Basin[J]. Xinjiang Petroleum Geology,2024,45(5):595-603.
• [8]伍秋姿，陈丽清，陈玉龙，等.基于机器学习算法的深部页岩储层物性预测及有利勘探区优选[J].非常规油气，2024,11(5):95-105.WU Qiuzi, CHEN Liqing, CHEN Yulong, et al. Prediction of physical properties of deep shale reservoirs and optimization of favorable exploration areas based on machine learning algorithms[J].Unconventional Oil&Gas,2024,11(5):95-105.
• [9]何佑伟，贺质越，汤勇，等.基于机器学习的页岩气井产量评价与预测[J].石油钻采工艺，2021,43(4):518-524.HE Youwei,HE Zhiyue,TANG Yong,et al. Shale gas well production evaluation and prediction based on machine learning[J]. Oil Drilling&Production Technology, 2021, 43(4):518-524.
• [10]谭茂金，白洋，张博栋.测井智能解释中机理模型-机器学习联合驱动范式及应用[J].石油地球物理勘探，2025,60(4):966-977.TAN Maojin, BAI Yang, ZHANG Bodong. Paradigms of joint driving by mechanistic model and machine learning in intelligent log interpretation and application[J]. Oil Geophysical Prospecting,2025,60(4):966-977.
• [11]冯国庆，杜勤锟，周道勇，等.基于机器学习的酸性气藏地下储气库硫化氢含量预测方法[J].天然气工业，2025,45(2):159-169.FENG Guoqing,DU Qinkun,ZHOU Daoyong,et al. Prediction of H2S content in sour gas reservoir underground gas storage based on machine learning[J]. Natural Gas Industry, 2025, 45(2):159-169.
• [12]柳洁，田冷，刘士鑫，等.基于复合机器算法的致密气井产能预测模型：以鄂尔多斯盆地SM区块为例[J].大庆石油地质与开发，2024,43(5):69-78.LIU Jie, TIAN Leng, LIU Shixin, et al. Prediction model of tight gas well productivity based on composite machine algorithm:Taking Block SM of Ordos Basin as an example[J].Petroleum Geology&Oilfield Development in Daqing, 2024, 43(5):69-78.
• [13]郑一鸣，李刚，季军，等.基于Stacking集成学习的盾构掘进地表沉降预测方法[J].隧道建设（中英文），2024,44(11):2233-2240.ZHENG Yiming,LI Gang,JI Jun,et al. Surface settlement prediction method for shield tunneling based on stacking ensemble learning[J]. Tunnel Construction, 2024, 44(11):2233-2240.
• [14]朱秋琳，梁艳萍，赵玉琴，等.基于机器学习的气井产气产水动态预测方法及应用[J].天然气技术与经济，2025,19(2):22-31.ZHU Qiulin, LIANG Yanping, ZHAO Yuqin, et al. A MLbased method to dynamically predict gas and water production in gas wells[J].Natural Gas Technology and Economy,2025,19(2):22-31.
• [15]张昀，陈彦润，陈晓刚，等.数据驱动的气井井筒积液与产量预测模型研究及应用[J].天然气与石油，2025,43(1):9-19.ZHANG Yun, CHEN Yanrun, CHEN Xiaogang, et al. Study and application of a data-driven prediction model for gas well liquid loading and production[J].Natural Gas and Oil,2025,43(1):9-19.
• [16]胡秋嘉，刘春春，张建国，等.基于机器学习的煤层气井产能预测与压裂参数优化[J].油气藏评价与开发，2025,15(2):266-273,299.HU Qiujia, LIU Chunchun, ZHANG Jianguo, et al. Machine learning-based coalbed methane well production prediction and fracturing parameter optimization[J]. Petroleum Reservoir Evaluation and Development,2025,15(2):266-273,299.
• [17]赵洪涛，李金泽，杨毅，等.基于长短期记忆神经网络的油井产液量和含水率预测方法[J].石油地质与工程，2022,36(5):75-80.ZHAO Hongtao,LI Jinze,YANG Yi,et al. Production and water cut performance predicting method based on long-short term memory neural network model[J].Petroleum Geology and Engineering,2022,36(5):75-80.
• [18]刘巍，刘威，谷建伟.基于机器学习方法的油井日产油量预测[J].石油钻采工艺，2020,42(1):70-75.LIU Wei, LIU Wei, GU Jianwei. Oil production prediction based on a machine learning method[J].Oil Drilling&Production Technology,2020,42(1):70-75.
• [19] VICENTE M, BELKENOFF R. Multi-regression modeling approach for choke performance and production analysis[C].Colorado:SPE/AAPG/SEG Unconventional Resources Technology Conference,2023.