摘要
参数拟合的传统页岩气井产能预测方法存在一定的局限性,引入基于支持向量机的非参数大数据分析方法进行页岩气井产能预测研究。根据生产数据记录以及井底压力随生产过程的变化规律,建立e-SVR支持向量回归模型,对长宁页岩气某区块实际生产数据分别进行了单井和多井的训练及预测检验,其中单井回归检验的相关度系数达到0. 958 556,体现了该方法优秀的回归能力;多井学习模型在前95 d区间内对单井数据的预测也达到了接近单井回归的效果,体现了该方法在密集数据区间内较好的预测能力,为页岩气产能预测提供了新的思路。
The traditional productivity forecast method of shale gas well based on parameter matching has some limitations.This paper introduced a non-parametric big data analysis method based on support vector machine(SVM).According to the recorded production data and the change law of bottomhole pressure with the production process,the e-SVR(support vector regression)model was established,and it was tested by single-well and multi-well training and forecasting based on actual production data of a block of Changning shale gas field.The correlation coefficient of single-well regression test reaches 0.958556,which reflects the excellent regression performance of this method.The forecasting of the multi-well learning model based on single well data in the first 95 d interval is close to the effect of the single-well regression model,which reflects the better forecasting performance of the method in the dense data interval.The proposed method provides a new idea for forecasting shale gas productivity.
作者
祝元宠
咸玉席
李清宇
卢徳唐
ZHU Yuanchong;XIAN Yuxi;LI Qingyu;LU Detang(Research Center of Petroleum, University of Science and Technology of China, Hefei, Anhui 230026, China)
出处
《油气井测试》
2019年第1期1-6,共6页
Well Testing
基金
国家科技重大专项"致密油气藏多尺度介质复杂结构井数值试井分析方法及应用研究"(2017ZX05009005-002)
中石油-中科院重大战略合作项目"页岩气钻完井井壁稳定与开发工程技术研究"(2015A-4812)
中国科学院战略先导科技专项"页岩气勘探开发基础理论与关键技术"(XDB10030402)
关键词
页岩气
产能预测
大数据
非参数拟合
支持向量机
长宁气田
shale gas
productivity forecast
big data
non-parametric matching
support vector machine(SVM)
Changning shale gas field
