页岩油压裂井产量预测方法研究  

Production prediction methods for fractured shale oil wells

作　　者：赵庆杰 唐宏宝 张乾 冯凡 郝华松 白石 ZHAO Qingjie;TANG Hongbao;ZHANG Qian;FENG Fan;HAO Huasong;BAI Shi(Downhole Technology Service Company,CNPC Bohai Drilling Engineering Company Limited,Tianjin 300280,China;Engineering Technology Department,CNPC Bohai Drilling Engineering Company Limited,Tianjin 300457,China;No.4 Brand Company of China Petroleum Pipeline Engineering Co.,Ltd,Langfang,Hebei 065000,China;Well Testing Company,CNPC Bohai Drilling Engineering Company Limited,Langfang,Hebei 065007,China)

机构地区：[1]中国石油集团渤海钻探工程公司井下技术服务公司,天津300280 [2]中国石油集团渤海钻探工程公司工程技术部,天津300457 [3]中国石油管道局工程有限公司第四分公司,河北廊坊065000 [4]中国石油集团渤海钻探工程有限公司油气井测试分公司,河北廊坊065007

出　　处：《油气井测试》2024年第3期46-52,共7页Well Testing

基　　金：中国石油集团渤海钻探工程有限公司重大技术研究项目“储层压裂改造‘数智化’技术研究”(2023ZD02F)。

摘　　要：为了综合考虑地质因素和工程参数对页岩油压裂井产量的影响,以大港油田页岩油井为数据来源,利用Spearman相关系数和随机森林综合筛选关键特征参数,通过数据清洗、多重填补检测剔除异常值,拓展缺失数据,构建完整的压裂井产量预测数据集。基于多层感知机神经网络模型,采用网格搜索法进行基础模型的参数调优,建立了页岩油压裂井产量预测模型,训练数据集的预测平均准确度为92.37%。经大港10口页岩油井的生产数据现场应用,预测产量与实际值的平均误差为7.59%,表明该产量预测模型可综合反应地质因素和工程参数对压裂井产量的影响,使预测结果与实际生产相吻合,预测精度高,满足工程需求。To comprehensively consider the impact of geological factors and engineering parameters on the production of fractured shale oil wells,by using the data from shale oil wells in the Dagang oilfield and by employing Spearman’s correlation coefficient and random forest techniques,key characteristic parameters were meticulously selected.Through processes of data cleansing,multiple imputation to detect and remove outliers,and expanding missing data,a complete dataset for predicting the production of fractured wells was constructed.Based on a multilayer perceptron neural network model,parameter tuning of the foundational model was achieved via grid search methodology,and a model for predicting the production of the fractured shale oil wells was established.The average prediction accuracy of the training dataset reached 92.37%.When applied to the production data from 10 shale oil wells in Dagang oilfield,the average error between predicted and actual production values was 7.59%.This indicates that the prediction model can effectively integrate geological and engineering parameters’influences on the production of fractured well,ensuring a high degree of alignment between predicted and actual production values,thereby meeting engineering requirements with high prediction accuracy.

关 键 词：页岩油 压裂 产量预测模型 多层感知机神经网络 现场试验 大港油田 特征参数 

分 类 号：TE353[石油与天然气工程—油气田开发工程]