仉文岗　李红蕊　巫崇智　王林

摘   要：针对传统地下采场开挖稳定评估方法存在的局限性，引入机器学习方法，提出基于随机森林算法（Random forest，RF）和K-最近邻算法（K-nearest neighbor，KNN）的地下采场开挖稳定性预测模型. 以加拿大8个采场为例，首先，获取并分析399组观测数据，其中涵盖了相应的岩石质量分级（Rock Mass Rating，RMR）值、跨度以及对应的稳定、潜在不稳定或不稳定状态. 然后将地下采场的稳定性程度进行三分类及二分类，采用10折交叉驗证方法进行模型超参数优化，在不作任何假设的前提下，捕捉地下采场开挖稳定性与RMR值、跨度之间的复杂关系. 研究表明：二分类结果准确性高于三分类预测结果;在二分类方式下，两种算法的准确率及召回率均高于90%，其中KNN算法的表现优于RF算法;提出的两种方法较先前研究的正确率有很大提升，为开挖稳定性评估提供了可靠途径.

关键词：随机森林;K-最近邻;开挖稳定性;交叉验证;召回率

中图分类号：TP181                            文献标志码：A

Stability Assessment of Underground Entry-type Excavations

Using Data-driven RF and KNN Methods

ZHANG Wengang1，2，3?，LI Hongrui1，WU Chongzhi1，WANG Lin1，2，3

（1. School of Civil Engineering，Chongqing University，Chongqing 400045，China;

2. National Joint Engineering Research Center of Geohazards Prevention

in the Reservoir Areas （Chongqing University），Chongqing 400045，China;

3. Key Laboratory of New Technology for Construction of Cities in

Mountain Area （Chongqing University），Ministry of Education，Chongqing 400045，China）

Abstract：In view of the limitations of traditional entry-type excavation stability assessment methods，this study explores the uses of novel data-driven machine learning methods to establish the excavation stability prediction models based on the random forest（RF） and K-nearest neighbor（KNN） methods. The proposed methods are based on 399 case histories from eight Canadian mines，covering a wide range of rock mass rating（RMR） and span，with stable，potential unstable and unstable cases categorized into ternary and binary groups. A ten-fold cross-validation method is applied to optimize the hyper-parameters during modeling. These two machine learning methods can capture the complex relationship between the excavation stability with RMR value and span without any assumptions of the underlying relationship. The results indicate that the accuracy of the binary classification results are slightly better than the ternary prediction results. For the binary classification circumstance，the accuracy and recall rate of both algorithms are higher than 90%，and the performance of the KNN algorithm is better than that of the RF algorithm. Meanwhile，the two proposed methods greatly improve the accuracy rate over previous studies and provide a reliable way for excavation stability assessment.

Key words：random forests;K-nearest neighbor;entry-type excavations stability;cross validation;recall rate

地下采场的稳定性是采矿作业安全性的主要关注点，对采场生产率有着重要影响. 确定采场稳定性的普遍方法是通过对岩石力学参数进行评估，然后采用岩石质量Q系统分类法、RMR （rock mass rating）分类法或地质强度指标（GSI）等方法对岩石进行评分，最终根据相应的分级标准[1-3]，确定岩石的稳定性程度. 近年来，一些方法如数值模拟法[4]、临界跨度图法[5]以及Mathews 稳定图法[6-7]等被提出用于预测矿洞的稳定性，然而这些方法比较傳统，经验性较强. 在数据挖掘时代，越……