一种消除测井数据多重共线性特征的密度测井数据重构技术及应用

在实际测量中，由于井下岩层的松散程度不一致，钻井技术限制，泥浆浸入，井眼条件差或者裂缝带的存在，造成井眼垮塌，从而无法获得高质量的测井曲线，特别是探测深度较浅的密度曲线。目前常用的密度重构方法为通过其他测井数据拟合密度，但该类方法一般存在2个方面问题：(1)拟合算法多为线性，实际资料中多为非线性关系；(2)测井曲线之间存在线性关系影响密度拟合精度。针对以上问题，首先利用井径和密度的交会图将地层分为标准层和校正层。然后分析了因变量(密度)与自变量(自然伽马、自然电位等常规测井数据)以及自变量相互之间的关系。其次利用核主成分分析去除自变量数据之间相关性，形成了相互正交的核主成分分量。最后选用随机森林非线性方法建立密度与核主成分分量关系模型，实际的密度重构效果在叠前地震勘探资料反演中取得了较好的应用效果。

A density logging data reconstruction technology and its application to eliminate multicollinearity of logging data

In the actual survey，due to the inconsistency of the loose degree of the underground strata，the limitation of the drilling technology，the mud immersion，the poor borehole conditions or the existence of the fracture zone，the borehole collapse was caused，and the high-quality logging curve，especially the density curve with a shallow detection depth，could not be obtained.At present，the commonly used density reconstruction method was to fit the density through other logging data，but this kind of method generally had two problems:on the one hand，the fitting algorithm was mostly linear，and the actual data was mostly nonlinear;On the one hand，there was a linear relationship between logging curves that affects the density fitting accuracy.In view of the above problems，the formation was divided into standard layer and correction layer by using the crossplot of well diameter and density.Then the relationship between dependent variable (density) and independent variable (natural gamma，natural potential and other conventional logging data) and independent variable was analyzed.Secondly，kernel principal component analysis was used to remove the correlation between independent variables and form mutually orthogonal kernel principal component.Finally，the random forest nonlinear method was used to establish the relationship model between density and kernel principal component，and the actual density reconstruction effect has achieved good application effect in pre-stack seismic exploration data inversion.