基于改进KNN算法的有限钻孔预测全域地质特征的方法

朱峻生, 王胜, 柏君, 徐正宣, 陈明浩, 李昭淇, 刘鑫, 张自豪, 刘兴倚

doi:10.3973/j.issn.2096-4498.2023.S2.039

隧道建设 >

2023 , Vol. 43 >Issue S2: 348 - 358

DOI: https://doi.org/10.3973/j.issn.2096-4498.2023.S2.039

地质与勘察

基于改进KNN算法的有限钻孔预测全域地质特征的方法

展开

（1. 成都理工大学地质灾害防治与地质环境保护国家重点实验室，四川成都 610059； 2. 中铁二院工程集团有限责任公司, 四川成都 610031； 3. 德阳市新源水利电力勘察设计有限公司，四川德阳 618000）

朱峻生（2002—），男，四川南充人，成都理工大学地质工程专业（钻掘方向）本科在读。E-mail： sacai@stu.cdut.edu.cn。*通信作者：王胜， E-mail： yongyuandewangsheng@sina.com。

网络出版日期: 2024-03-28

收起

An Improved KNearest Neighbor Algorithm Method Using Finite Boreholes for Predicting FullArea Geological Features

Expand

(1. State Key Laboratory of Geohazard Prevention and Geo-environment Protection, Chengdu University of Technology, Chengdu 610059, Sichuan, China; 2. China Railway Eryuan Engineering Group Co., Ltd., Chengdu 610031, China; 3. Water Conservancy and Electric Power Survey and Design Co., Ltd., Deyang 618000, Sichuan, China)

Online published: 2024-03-28

Fold

摘要

为解决传统建模方法难以对原始钻孔数据进行深度挖掘和利用的问题，提出一种改进KNN（K-nearest neighbor）算法，它是在原有的KNN算法的基础上，通过实现k值根据不同地层自动选取以及对原始地质资料的进一步利用，形成的一套空间自适应插值拟合算法。将一铁路勘查工程数据作为数据源，导入该数据到改进KNN算法模型中并运行，成功获得了该地各地层的特征k值，并实现了地质建模。通过对比实际建模以及实际的钻孔柱状图与KNN改进前后预测的钻孔柱状图发现，改进KNN算法对于薄层的预测更加准确，总体准确率更高。并且通过机器学习指标的对比验证发现，相比于原始KNN算法以及其他常见分类算法，改进KNN算法能够获得更好的地层预测效果，做到了“求全”与“求精”，能够较好地指导地下三维空间的预测。

关键词： 机器学习; K邻近域算法; 全域地质特征预测; 地质建模; 钻孔数据

本文引用格式

朱峻生, 王胜, 柏君, 徐正宣, 陈明浩, 李昭淇, 刘鑫, 张自豪, 刘兴倚 . 基于改进KNN算法的有限钻孔预测全域地质特征的方法[J]. 隧道建设, 2023 , 43(S2) : 348 -358 . DOI: 10.3973/j.issn.2096-4498.2023.S2.039

Abstract

The original drilling data collected cannot be deeply excavated and utilized by traditional modeling methods. Therefore, an improved Knearest neighbor(KNN) algorithm is improved. This is a spatial adaptive interpolation fitting algorithm that is developed based on the original KNN algorithm. It incorporates automatic selection of 〖WT5BX〗k〖WT〗〖WT５《TNR》〗values based on different geological layers and utilizes the original geological data for further analysis. The data from a railway investigation project is selected as a data source to put into the improved KNN algorithm model, successfully obtaining the characteristic 〖WT5BX〗k〖WT〗〖WT5《TNR》〗values of each stratum in the site and accomplishing the geological modeling. By comparing the actual borehole and the borehole predicted by original KNN and improved KNN, it is found that the improved KNN algorithm is more accurate in predicting the thin layers, and the overall accuracy is higher. Compared with the original KNN algorithm and other common classification algorithms, the improved algorithm can obtain better stratum prediction results, especially for the thin stratum, and obtain higher precision and accuracy, which can better guide the prediction of underground 3D space.

Key words： machine learning; Knearest neighbor; prediction of fullarea geological features; geological modeling; drilling data

Options

文章导航

〈

〉

模态框（Modal）标题

摘要

本文引用格式

Abstract