摘要
最小二乘支持向量机是在统计学习理论基础上发展起来的模式识别方法。与传统统计学相比,它能有效解决有限样本、非线性、高维数模型的建立问题,而且建立的模型具有很好的预测性能。岩性识别本质是解决分类问题,本文基于最小二乘支持向量机解决分类问题的优势,首先用GR、CNL、DEN、AC、RLLD等常规测井曲线数据建立样本空间;然后通过耦合模拟退火和交叉验证的方法寻找最佳参数,优化最小二乘支持向量机分类器;最后建立了最小二乘支持向量机岩性识别模型。通过取心段岩心描述和岩心/岩屑薄片鉴定,确定辽河盆地40口井315m井段2 520个岩性样品作为训练样本,建立岩性识别标准。对8口井13 866m井段110 928个火山岩数据采样点进行测井识别,可识别致密玄武岩、气孔玄武岩、粗面岩等8种主要火山岩类型。识别结果与8口测试井中316个有取心段岩心描述和岩心/岩屑薄片的精确岩矿定名对比,符合率达到75.2%,与以往测井识别复杂火山岩岩性相比,在识别准确率和效率上都有明显提高。
Least squares support vector machine （LS-SVM ） is a pattern recognition method developed from the statistical learning theory.Compared with the traditional statistics,LS-SVM can＆amp;nbsp;effectively resolve the problems of the finite of samples,non-linearity and high-dimension,and it can achieve accurate prediction.The essence of lithology identification is for classification.We take the advantage of the classification by LS-SVM：at first,the sample space is established by using the conventional logging curves of GR、CNL、DEN 、AC、R LLD;and then,the classifier of LS-SVM is optimized by searching optimal parameters using the simulated annealing algorithm and cross validation method;finally,the model of LS-SVM lithology identification is determined.Through the description of core section and the analysis of core/cuttings,2 520 samples from the 315 m section of the 40 wells are taken as training samples in Liaohe basin,for establishing the standard of lithology identification. 1 10 928 logging data from 13 866 m section of the 8 wells are taken as the predicting samples.8 types of volcanic rocks have been identified such as vesicular basalt,compact basalt,trachyte, et al. In comparison with the 31 6 samples from 8 wells,the corresponding identification rate is 75.2%.The accuracy and velocity of the LS-SVM lithology identification is improved distinctly compared with other well logging methods.
出处
《吉林大学学报（地球科学版）》
EI
CAS
CSCD
北大核心
2015年第2期639-648,共10页
Journal of Jilin University：Earth Science Edition
基金
国家"973"计划项目(2012CB822002)
