基于机器学习和编码器耦合的滑坡易发性评价  

Landslide susceptibility assessment based on machine learning and encoder coupling

作　　者：张萌萌 李少达[1] 王潇 李欣玥 戴可人 ZHANG Mengmeng;LI Shaoda;WANG Xiao;LI Xinyue;DAI Keren(College of Earth and Planetary Science,Chengdu University of Technology,Chengdu 610059,Sichuan,China;School of Architecture and Civil Engineering,Chengdu University,Chengdu 610106,Sichuan,China;Mahindra United World College of India,Pune,MH 412108,India;State Key Laboratory of Geohazard Prevention and Geoenvironment Protection,Chengdu University of Technology,Chengdu 610059,Sichuan,China)

机构地区：[1]成都理工大学地球与行星科学学院,四川成都610059 [2]成都大学建筑与土木工程学院,四川成都610106 [3]印度马轩德拉世界联合学院,印度浦那412108 [4]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

出　　处：《河南理工大学学报(自然科学版)》2025年第2期69-80,共12页Journal of Henan Polytechnic University(Natural Science)

基　　金：国家自然科学基金资助项目(41801391)。

摘　　要：目的为了在有限样本下提升机器学习模型挖掘数据特征的能力,提高模型预测精度,方法选取四川省雅砻江和大渡河中下游省级水土流失重点预防区九龙县、康定市、泸定县和木里县为研究区,选择12个影响因子构建滑坡易发性评价指标体系,使用确定性系数(certainty factor,CF)量化评价指标,对比逻辑回归(logistic regression,LR)和支持向量机(support vector machine,SVM)模型,在表现最优的模型上添加降噪自编码器(denoising autoencoder,DAE)和卷积自编码器(convolutional auto-encoders,CAE),并对比各模型所提取数据的特征。结果结果表明:CF-SVM模型的精确率(P),F-measure、Kappa系数,总准确度(overall accuracy,OA)和ROC曲线下与坐标轴围成的面积(area under curve,AUC)相较于CF-LR模型的分别提高了31.9%,1.1%,17.1%,8.5%,8.6%;添加DAE编码器后,CF-SVM-DAE模型的召回率(R),Fmeasure,Kappa系数和总准确度(OA)相比于CF-SVM模型的分别提高了8.1%,5.8%,8.1%,4%;添加CAE编码器后,CF-SVM-CAE模型的召回率(R),F-measure,Kappa系数和总准确度(OA)相比于CF-SVM模型的分别提高了0.4%,0.2%,0.2%,0.1%。结论选用的机器学习方法中,CF-SVM模型预测精度更高。在CF-SVM模型基础上添加DAE编码器比添加CAE编码器鲁棒性更好,因此,CF-SVM-DAE模型在所有模型中表现最好,更适合滑坡易发性评价。Objectives To enhance the ability of machine learning models to extract data features with lim-ited samples and improve the predictive accuracy of the models,Methods Jiulong County,Kangding City,Luding County and Muli County,the key provincial erosion prevention areas in the middle and lower reaches of the Yalong and Dadu Rivers in Sichuan Province,were selected as the study area for landslide susceptibility evaluation.Twelve influencing factors were selected to construct the landslide susceptibility evaluation index system,the coefficient of determination(CF)was used to quantify the evaluation index,and noise-reducing auto-encoders(DAEs)and convolutional auto-encoders(CAE)were added to the best-performing model by comparing the logistic regression(LR)and the support vector machine(SVM)models.Results The results showed that compared with the CF-LR model,the CF-SVM model,the precision(P),F-measure,Kappa coefficient,overall accuracy(OA),and AUC of the CF-SVM model increased by 31.9%,1.1%,17.1%,8.5%,and 8.6%,respectively,After adding the DAE encoder,the recall(R),F-measure,Kappa coefficient,and overall accuracy(OA)of the CF-SVM-DAE model increased by 8.1%,5.8%,8.1%,and 4%,respectively,compared to the CF-SVM model After adding CAE encoders,the re-call(R),F-measure,Kappa coefficient,and overall accuracy(OA)of the CF-SVM-CAE model increased by 0.4%,0.2%,0.2%,and 0.1%,respectively,compared to the CF-SVM model.Conclusions The CF-SVM model has higher prediction accuracy among the selected machine learning methods.Adding DAE en-coder toto the CF-SVM has better robustness than adding CAE encoder,thus the CF-SVM-DAE model per-forms the best among all models and is more suitable for the current study area.

关 键 词：滑坡 易发性建模 LR SVM 编码器 

分 类 号：P642.22[天文地球—工程地质学]