基于Fisher判别分析法的流体识别

A method of fluid recognition basing on Fisher discriminant analysis

Fisher判别分析法的基本方法是将多维数据点投影到一条直线上。然后按方差分析的思想选出最优投影方向,使得投影后样品总体包含的各种类型能尽可能分开。使用Fisher判别法进行油气、水层的识别,首先要选择能反映油气、水层特点且尽可能独立的测井曲线,建立样品点的观测向量;然后根据己有试油资料或确有把握的油气层与水层的已知数据,分别建立油气层和水层的统计总体;再设立判别向量,通过对样品变量作变换而将其投影于判别向量方向,建立判别函数F;此后,将已知油气层和水层的数据作为样本代入判别函数,根据Fisher准则(两类样本总体之间差距最大而每类样品内的离散性最小)求取其中的待定系数,从而确定判别函数。

The basic method of Fisher discriminant analysis is as follows.The multidimension- al data points are projected on a straight line firstly, then based on the theories of analysis of variance,the optimum projection direction is picked out,so that the various types of samples can be separated as possible.Fisher discriminant analysis can effectively identify oil and gas reservoir.First of all,the selected logging curve should reflect the characteristics of oil,gas and water layer and be independent as possible,meanwhile, the observation vector of the sample points are set up.After that,according to the test data or the data of oil, gas layer and water layer, the statistical population of oil,gas layer and water layer are set up separately. Then, the discriminant vectors are determined,and through a transformation of the sample variables, they are projected to discriminate the vector direction and establish the discriminant function F.After that,as samples,the data of the oil,gas layer and water layer are generated into the discriminant function.According to the Fisher ratio, namely, greatest difference exists between two types of samples in total,while the minimum discreteness exists within each type of samples,the undetermined coefficients are worked out, so as to the discriminant function is determined.