基于Lasso-Lars的密井网水驱注水量影响因素分析  

Research on Water Injection Rate of Dense Well Pattern using Lasso-Lars

作　　者：王梅[1,2] 孙莺萁 宋考平[2,3] 高雅田[1] 杨二龙[3] 

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318 [2]北京德威佳业科技有限公司博士后工作站,北京300072 [3]东北石油大学提高采收率教育部重点实验室,黑龙江大庆163318

出　　处：《数学的实践与认识》2016年第8期124-131,共8页Mathematics in Practice and Theory

基　　金：黑龙江省科学基金(F2015020);北京市博士后工作经费资助项目;北京市朝阳区博士后工作经费资助项目;东北石油大学校培育基金项目(XN2014102)

摘　　要：目前大部分油藏工程都需要合理的注水量调整方案,为了准确预测注水量则需要分析注水量的影响因素及其之间的关系。通过lasso方法可将模型的系数进行压缩使之变小趋于0,利用lars算法可有效解决lasso的求解问题并记录正则化参数λ所有可能取值下对应的lasso优化问题的解,求得lasso正则化路径.应用lasso-lars正则化路径,得到每一个注水井注水量影响因素对应的回归系数及回归系数变化走势图,确定不同影响因素对注水井注水量的敏感程度.同时证明该方法相对于其他方法的有效性及优越性,对注水量预测模型的建立具有重要意义.As most of the reservoir engineering needs the water injection adjustment programs,it is necessary to analyze the relation of the influence factors.By the method of lasso,model coefficient can be reduced to zero.The lars algorithm can be used effectively to solve the lasso problems and record all the possible values of the lasso solutions of optimization problem,order to obtain the lasso regularization path.Using the lasso-lars regularization path to obtain the regression coefficient and regression coefficient variation of each water injection,and the sensitivity of different factors to water injection was determined.At the same time,it is proved that the method is effective and superior to other methods,and it also has great significance for the establishment of water injection rate prediction model.

关 键 词：Lasso LARS 正则化路径 密井网 注水量 影响因素 

分 类 号：TE357.6[石油与天然气工程—油气田开发工程]