基于机器学习的边坡稳定性分析方法——以国内618个边坡为例  

Slope Stability Analysis Method Based on Machine Learning—Taking 618 Slopes in China as Examples

作　　者：张梦涵 魏进[1] 卞海丁 ZHANG Meng-han;WEI Jin;BIAN Hai-ding(School of Highway,Chang an University,Xi'an 710064,Shaanxi,China)

机构地区：[1]长安大学公路学院,陕西西安710064

出　　处：《地球科学与环境学报》2022年第6期1083-1095,共13页Journal of Earth Sciences and Environment

基　　金：陕西省交通厅科技项目(22-24K);西安市建设科技计划项目(SZJJ2021-10);中央高校基本科研业务费专项资金项目(310821173701);国家自然科学基金项目(52178310)。

摘　　要：为快速精确预测边坡稳定性状态,提出了一种基于机器学习的边坡稳定性状态智能评估方法。基于边坡失稳特征,结合国内618个边坡案例,选取了6个典型边坡参数——重度(γ)、黏聚力(C)、内摩擦角(φ)、坡角(β)、坡高(H)和孔隙水压力(P),建立了边坡稳定性评价数据集。采用机器学习理论中的梯度提升机(GBM)、支持向量机(SVM)、人工神经网络(ANN)及随机森林(RF)算法分别建立边坡稳定性预测模型,利用训练集对模型训练学习,使用五折交叉验证和网格搜索法对模型进行参数调整,并开展精度评价。基于受试者工作特征曲线下面积(AUC)和F_(1)分数(F_(1)Score)可知,随机森林算法的AUC值为0.969,F_(1)分数为0.904,随机森林算法的评价指标最优,更适合用于分析边坡稳定性。基于随机森林算法分别删除不同特征变量建立的不同边坡稳定性预测模型,得到特征参数敏感程度从大到小为重度、坡角、坡高、内摩擦角、孔隙水压力、黏聚力,并基于特征参数敏感程度提出了针对不同敏感因素的边坡防护措施。In order to quickly and accurately predict the slope stability,an intelligent evaluation method of slope stability based on machine learning was proposed.Based on the characteristics of 618 slopes in China,6 typical slope parameters,including gravity(γ),cohesion(C),internal friction angle(φ),slope angle(β),slope height(H)and pore water pressure(P),were selected to establish the slope stability evaluation dataset.Slope stability prediction models were established using gradient boosting machine(GBM),support vector machine(SVM),artificial neural network(ANN)and random forest(RF)in the machine learning algorithm.The models were trained and learned using the training set.The model parameters were adjusted using the 5-fold cross validation and grid-search.The ability of the model to classify the slope stability was tested through the test set,and the optimal model was determined.The results show that by comparing the area under the receiver operating characteristic curve(AUC)and F_(1)Score of GBM,SVM,ANN and RF algorithms,the AUC value and F_(1)Score of RF algorithm are 0.969 and 0.904,respectively;RF algorithm has the best evaluation index and is more suitable for analyzing slope stability.Based on the different slope stability prediction models established by deleting different characteristic variables of RF algorithm,the characteristic parameters in descending order of sensitivity areγ,β,H,φ,P,C.According to the different sensitivity factors,the slope protection measures are proposed based on the sensitivity of the characteristic parameters.

关 键 词：边坡 稳定性 机器学习 随机森林 智能评估 特征参数 混淆矩阵 防护措施 

分 类 号：P642.2[天文地球—工程地质学] TU457[天文地球—地质矿产勘探]