摘要
准确预测页岩体积压裂井的产能是确定合理开发决策的重要前提。目前页岩气井产能预测主要基于理论模型,需要理想化假设条件和不易得到的参数,导致体积压裂前的产量预测精度不高。为此,通过数据挖掘技术直接从影响产能的参数入手,突破传统理论模型的局限,首先利用灰色关联度确定影响长宁地区57口页岩气水平井压后产量的主控因素及权重,然后基于遗传算法优化的误差反向传播(back propagation,BP)神经网络方法,建立页岩气水平井体积压裂产能预测模型。基于该模型,针对长宁地区已生产井数据开展现场应用。应用结果表明:工程参数主要影响页岩气水平井的初期产量,总有机碳含量(total organic carbon,TOC)、单井百米液量、单井百米砂量、脆性矿物指数等工程参数是影响页岩气水平井测试产量和3个月累产气量的主控因素;TOC、I类储层钻遇长度、孔隙度、含气量等地质参数是影响页岩气水平井1年累产气量的主控因素;基于长宁地区已生产井数据建立的页岩气水平井体积压裂测试产量预测模型的平均误差为8.76%,预测误差同比多元回归模型预测降低了47.79%;基于遗传算法-误差反向传播(genetic algorithm-back propagation,GA-BP)神经网络的产能预测技术具有操作灵活和预测精度高的特点。利用大数据分析和产能预测方法为长宁地区页岩气井的产能预测提供一种新思路,提高了产能预测效率,并有效地指导现场施工。
Accurately predicting the volumetric fracturing capacity of shale gas horizontal wells is an important prerequisite for the decision-making of the rational development.At present,productivity prediction of shale gas horizontal wells is mainly based on theoretical models,which require idealized assumptions and hard-to-obtain parameters,resulting in low accuracy of model prediction.Therefore,started with the data mining technology directly from the parameters affecting the production capacity,breaking through the limitations of the traditional theoretical model,by using the gray correlation degree,the main controlling factors and weights affecting the post-pressure production of 57 shale gas horizontal wells in Changning area were determined.Then based on back propagation(BP)neural network method optimized by genetic algorithm,a volumetric productivity of fractured horizontal wells in shale gas reservoirs model was established.Thin model was applied on the scene to the production well data in Changning area.The application results show that,The engineering parameters mainly affect the initial production of shale gas horizontal wells.The engineering parameters such as TOC,single well 100 meters liquid volume,single well 100 meters sand volume,and brittle mineral index and other engineering parameters are the main controlling factors affecting the shale gas horizontal well test yield and 3 months of gas production.The geological parameters such as TOC,I reservoir drilling length,porosity and gas content are the main controlling factors affecting the gas production of shale gas horizontal wells for one year.Based on the production well data in Changning area,the average error of the shale gas horizontal well volume fracturing test yield prediction model is 8.76%,and the prediction error is reduced by 47.79%compared with the multivariate regression model.Production prediction technology based on GA-BP neural network has the advantages of flexible operation and high prediction accuracy.The data mining based analysis method provides a new idea for productivity prediction of shale gas horizontal wells in Changning area.It can improve the productivity prediction efficiency,and effectively guide on-site fraturing construction.
作者
陈娟
黄浩勇
刘俊辰
曾波
杨昕睿
CHEN Juan;HUANG Hao-yong;LIU Jun-chen;ZENG Bo;YANG Xin-rui(Shale Gas Research Institute of Southwest Oil and Gas Field Company,Chengdu 610017,China;Sichuan Shale Gas Company of Southwest Oil and Gas Field Company,Chengdu 610017,China)
出处
《科学技术与工程》
北大核心
2020年第5期1851-1858,共8页
Science Technology and Engineering
基金
国家重大科技专项(2016ZX05023-005-003)。
关键词
页岩气
神经网络
遗传算法
产能预测
大数据分析
shale
neural network
genetic algorithm
productivity prediction
big data analysis
作者简介
第一作者:陈娟(1985—),女,汉族,四川崇州人,硕士,工程师。研究方向:页岩气储层改造技术。E-mail:c_juan@petrochina.com.cn。
