基于机器学习的页岩气产能非确定性预测方法研究  

Non-Deterministic Shale Gas Productivity Forecast Based on Machine Learning

作　　者：马文礼[1,2] 李治平 孙玉平[3] 张静平 邓思哲[4] Ma Wenli;Li Zhiping;Sun Yuping;Zhang Jingping;Deng Sizhe(China University of Geosciences (Beijing), Beijing 100083, China;Beijing Key Laboratory of Unconventional Natural Gas Geology Evaluation and Development Engineering,China University of Geosciences (Beijing), Beijing 100083, China;PetroChina Research Institute of Petroleum Exploration and Development, Langfang, Hebei 065007, China;China Volant Industry CO., LTD, Beijing 100080, China)

机构地区：[1]中国地质大学(北京),北京100083 [2]非常规天然气能源地质评价与开发工程北京市重点实验室,北京100083 [3]中国石油勘探开发研究院,河北廊坊065007 [4]中国华腾工业有限公司,北京100080

出　　处：《特种油气藏》2019年第2期101-105,共5页Special Oil & Gas Reservoirs

基　　金：国家科技重大专项"页岩气开发规模预测及开发模式研究"(2016ZX05037-006)

摘　　要：针对页岩气确定性产能预测方法误差较大的问题,综合最大信息系数相关性分析方法、混合支持向量机技术及"蒙特卡洛-马尔科夫链"模拟,提出一种基于机器学习的页岩气产能非确定性预测方法。运用该方法,可根据已投产页岩气井的地质及工程数据,对拟钻页岩气井未来的产能进行非确定性预测。24口页岩气井算例分析结果表明:利用该方法进行产能非确定性预测的准确率为70. 8%,且预测结果为"大概率事件"的井占54. 2%,说明该方法有较高的预测精度且预测结果满足概率统计规律。研究成果对国内外页岩气开发方案的优化有重要意义。In order to improve the deterministic shale gas productivity, forecast precision, a non-deterministic shale gas productivity forecast method based on machine learning was proposed by combining with maximum information coefficient correlation analysis, hybrid support vector machine and “Monte Carlo-Markov chain” simulation. Based on the geological and engineering data of available shale gas wells had been put into production, this method can be used to forecast the future productivity of shale gas well to be drilled. Illustrative application of 24 shale gas wells shows that the forecast accuracy for this non-deterministic shale gas productivity is 70.8%, and the number of wells with “high probability event” accounts for 54.2%. This new productivity forecast method with relative high accuracy can meet probability statistics. This research could provide certain significance for the shale gas development program optimization at home and abroad.

关 键 词：产能预测 非确定性 页岩气 混合支持向量机 蒙特卡洛-马尔科夫链 机器学习 

分 类 号：TE349[石油与天然气工程—油气田开发工程]