基于机器学习的高寒山区矿山地质灾害易发性研究

Study on the mining geological hazard susceptibility assessment in alpine areas using machine learning models

高寒山区矿山地质环境脆弱,受冰川活动和人类工程活动影响,地质灾害频发,严重制约矿山的高效开采和安全运营.地质灾害易发性研究是矿山防灾减灾和科学预警的必要手段.本研究以天山诺尔湖铁矿为例,采用J48决策树、逻辑模型树(LMT)和径向基函数分类器(RBFC)为代表的机器学习模型来预测高寒山区矿山地质灾害易发性.基于地质图、高分辨率遥感影像和现场调查,准备256个地质灾害点和12个孕灾环境因子建立地质灾害数据集,并将地质灾害随机划分为训练集(80%)和验证集(20%);使用多重共线性诊断和信息增益进行因子的特征选择;最后利用受试者工作特征曲线下的面积(A_(UC))和统计指数(如Accuracy,简写为A_(cc))对比地质灾害易发性分区结果及模型的预测性能.结果显示:坡度、距道路距离和距河流距离是矿区地质灾害发生的主控因子.RBFC模型的预测能力最高(A_(UC)=0.854,A_(cc)=79.41%),其次是LMT(A_(UC)=0.849,A_(cc)=77.45%)和J48(A_(UC)=0.819,A_(cc)=75.49%).地质灾害高和极高易发区多分布于距道路和河流距离<1000 m、距矿山工程和冰川距离<2000 m的区域,3种模型生成的地质灾害易发性分区图均能表现矿山实际地质灾害分布状况.本研究可为矿山工程的选址以及土地利用规划提供科学参考,提高矿山地质安全评估及应急管理能力.

The fragile geological environment of mines in alpine regions,influenced by glacier activity and human engineering,experiences frequent geological hazards that significantly hinder efficient mining and safe operations.Assessing geological hazard susceptibility has become essential for disaster prevention and mitigation,and scientific early warning in mining areas.This study examined the Nuoer Lake iron mine in Tianshan,employing machine learning models such as the J48decision tree,logical model tree(LMT),and radial basis function classifier(RBFC)to predict geological hazard susceptibility in alpine mining areas.Initially,ageospatial database for the study area was created using geological maps,high-resolution remote sensing images,and on-site investigations.This database included 256geological hazards and 12conditioning factors,which were randomly split into training(80%)and validation(20%)datasets.Subsequently,geological hazard conditioning factors were selected through multicollinearity diagnosis and information gain.Finally,the performance of the models was evaluated and validated using the area under the receiver operating characteristic curve(A_(UC))and statistical indices(e.g.,Accuracy,abbreviated as A_(cc)).The results indicate that slope angle,proximity to roads,and proximity to rivers are the most significant parameters influencing geological hazards.The RBFC model attains the highest prediction accuracy(A_(UC)=0.854,A_(cc)=79.41%),followed by the LMT(A_(UC)=0.849,A_(cc)=77.45%)and J48(A_(UC)=0.819,A_(cc)=75.49%)models.Geological hazard high and very high susceptibility zones are primarily located within 1000meters of roads and rivers,and within 2000meters of mining projects and glaciers.The geological hazard susceptibility maps generated by all three models accurately reflect the actual distribution of geological hazards in mining areas.This study provides scientific references for site selection and land-use planning in mining engineering,enhancing geological safety assessment and emergency management capabilities.