首页|基于机器学习方法的测井岩相分类研究

基于机器学习方法的测井岩相分类研究

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
目前,在基于测井数据的岩相分类研究中,机器学习算法是一个研究热点.然而,实际上钻井取心极少,岩相样本匮乏,机器学习算法就会遇到过拟合问题,从而导致测井岩相分类效果不佳.因此,本文研究小样本环境下不同机器学习算法在测井岩相分类工作中的预测效果.以北美Panoma油气田数据集为例,通过逐步减少训练样本数量,建立了四种训练模式.同时,选用三类有代表性的有监督学习算法,以评估小样本环境下不同算法的预测效果,包括基于一般梯度下降的线性回归分类算法、支持向量机算法和一维卷积神经网络算法.在综合评价算法的预测效果时,选用了岩相分类准确率、总体岩相分类F1值、各岩相分类F1值以及有效识别最大相数.结果表明,随着训练样本数量的减少,三种算法的预测效果并未呈现线性下降趋势,且一维卷积神经网络较另外两种算法表现更为稳健.
Study on well-log lithofacies classification based on machine learning methods
Machine learning algorithms have been widely applied in the study of lithofacies classification based on well-log data at present.However,with very few drilled cores and a scarcity of lithofacies samples,machine learning algorithms encounter overfitting problems.This leads to poor well-log lithofacies classification.Therefore,the prediction effectiveness of different machine learning algorithms were studied in small sample environment,taking the North American Panoma oil and gas field dataset as an example.Four training models are established by gradually reducing the number of training samples.Meanwhile,three different types of supervised learning algorithms were selected to evaluate the prediction effectiveness,including Linear Regression Classification based on General Gradient Descent(GGD-LRC),Support Vector Machine(SVM),and One-Dimensional Convolutional Neural Networks(1D-CNN).In the comprehensive evaluation of the prediction effectiveness of the algorithms,selected the lithofacies classification accuracy,the overall lithofacies classification Fl value,the individual lithofacies classification Fl value,and the maximum number of effectively identified lithofacies.The results show that the prediction effectiveness of the three algorithms does not present a linear downward trend as the number of training samples decreased,and the 1D-CNN algrithm is more robust than the others.

Well-log lithofacies classificationMachine learningLinear regression classificationSupport Vector Machine(SVM)One-Dimensional Convolutional Neural Networks(1D-CNN)

高飞、曲志鹏、魏震、朱剑兵、程远锋

展开 >

新疆大学地质与矿业工程学院,乌鲁木齐 830017

中国石化胜利油田物探研究院,东营 257000

新疆大学中亚造山带大陆动力学与成矿预测实验室,乌鲁木齐 830017

测井岩相分类 机器学习 线性回归分类 支持向量机 一维卷积神经网络

新疆维吾尔自治区基金项目新疆维吾尔自治区"天池英才"引进计划项目

2022D01C422

2024

10.6038/pg2024HH0065

地球物理学进展
中国科学院地质与地球物理研究所 中国地球物理学会

地球物理学进展

CSTPCD北大核心
影响因子：1.761
ISSN：1004-2903
年,卷(期)：2024.39(3)