基于随机森林的地质解释方法与实现-计算机系统结构专业论文.docxVIP

摘 要 本文利用数据挖掘技术提出一种基于随机森林分类器对油田测井数据地质解释的 方法。利用数据挖掘分类方法构建地质解释分类器模型，对测井数据进行分类解释，是 一个低成本的泛化的地质解释手段，对石油勘探与开发具有显著的决策意义。论文通过 对大量的文献的研究，探讨通过 R 语言构建测井样本数据随机森林分类器模型、评估 与实现随机森林分类器的方法。油井测井数据变量维度高、数据量大、类型复杂，是一 种典型的“多元分类”问题。讨论两种主要的划分子集方法，方法一以样本特征变量为 主划分样本；方法二以样本解释类别为主，利用“二元分类”方法的思想对样本进行划 分，分割成多个样本子集。以方法二作为用于地质解释的随机森林建模样本集的划分方 法，建模用的样本集以“可能含油气”的解释类别为主要样本构成。为改善样本集建模 存在“过拟合”，在样本集中添加小量“噪声”样本数据，进一步提高随机森林分类器 准确率。提出的解释类别为主的样本子集划分方法，是解决“多元分类”问题的一个高 效的方法之一。 关键词：分类器 随机森林 数据挖掘 地质解释 测井数据 Abstract A classification approach based on Random Forests of Data Mining techniques to obtain the class of geologic interpretation from logging data is presented in this article. It has significant meaning for the decision in petroleum exploration and development to model the classifier depended on the geologic interpretation samples to obtain a new low-cost and generalized interpretation in new logging dataset. According to the latest researches, an approach for modeling, evaluation and deployment Random Forest classifier of geologic interpretation in R language is probe in the paper. The logging has the characteristics of high variable dimension, a large quantity of data, complex types. It is a typical multi-label issuse. Two approaches of dividing subsets are discussed. One is depended on the variable features; the other is based on the class. The latter refer to binary relevance method for multi-label classification. The sample subset is basen on the “possible oil and gas” class to divide. In order to avoid over fitting emerge as the sample set modeling classifier; some “noise” are appended into set. The accuracy of the Random Forests Classifier is improved further. The method can be applied to solve multi-label classification issues well. Keywords: Classifier Random Forests Data Mining Geologic Interpretation Logging Data 目 录 HYPERLINK \l _bookmark0 第一章 绪论 1 HYPERLINK \l _bookmark1 1.1 研究背景 1 HYPERLINK \l _bookmark2 1.2 研究动机 1 HYPERLINK \l _bookmark3 1.3 研究目的 2 HYPERLINK \l _bookmark4 1.4 论文工作及内