基于支持向量机的深层地应力预测模型  

SVM deep geo-stress prediction model

作　　者：赵小龙[1] 

机构地区：[1]中国石化胜利油田分公司,山东东营257000

出　　处：《特种油气藏》2016年第1期139-141,158,共3页Special Oil & Gas Reservoirs

基　　金：国家科技重大专项"胜利油田薄互层低渗透油田开发示范工程"(2011ZX05051)

摘　　要：为解决工程中利用测试或计算的少量地应力资料来反演计算深井地层应力场问题,运用建立在统计学习理论基础上的支持向量机(SVM)理论建立计算模型,利用井场实测的水平最大地应力和孔隙压力场数据进行了拟合,并对拟合计算中核函数的影响进行了对比分析。研究结果表明:SVM预测模型所预测的水平最大地应力值相对真实值的最大误差为5.1%,所预测的孔隙压力值相对真实值的最大误差为1.553%,模拟结果误差较小;与滑动最小二乘法的模拟结果相对比,SVM预测值与实测数据吻合度较高;拟合计算中RBF核函数计算精度较好。研究结果表明,基于支持向量机的深层应力场预测模型的精度可满足工程需求。A small amount of test or calculated data is usually used to inversely calculate deep- well geo- stress field in engineering field. The SVM( Support Vector Machine) theory based on statistics is used to establish a calculation model. The field test data of maximum horizontal geo- stress and pore- pressure are used to match and analyze the effect of kernel function. Result illustrates that there are relatively small errors between predicted value of SVM model and real value,and the maximum errors in maximum horizontal geo- stress and pore- pressure are 5. 1% and 1. 553% respectively. The SVM prediction is perfectly matched with field test data comparing with the prediction of moving least squares method. RBF kernel function remains a higher accuracy in the prediction,which indicates that this SVM geo- stress prediction model is applicable to engineering application.

关 键 词：地应力 孔隙压力 支持向量机 回归预测 核函数 

分 类 号：TE319[石油与天然气工程—油气田开发工程]