基于聚类分析和支持向量机的滑坡易发性评价  

Landslide susceptibility assessment based on clustering analysis and support vector machine

作　　者：黄发明[1] 殷坤龙[1] 蒋水华[2] 黄劲松[2,3] 曹中山 

机构地区：[1]中国地质大学(武汉)地质调查研究院,湖北武汉430074 [2]南昌大学建筑工程学院,江西南昌330000 [3]纽卡斯尔大学岩土科学与工程卓越研究中心

出　　处：《岩石力学与工程学报》2018年第1期156-167,共12页Chinese Journal of Rock Mechanics and Engineering

基　　金：国家自然科学基金面上项目(41572292;51509125;51679117)~~

摘　　要：在将支持向量机(support vector machine,SVM)等机器学习模型用于区域滑坡易发性评价时,大都随机或主观地选取非滑坡栅格单元,不能保证所选的非滑坡栅格单元是真正的"非滑坡"。为解决此问题,提出基于聚类分析和SVM的滑坡易发性评价模型。该模型首先用自组织映射(self-organizing mapping,SOM)神经网络对滑坡易发性进行聚类分析;然后从极低易发区中选择非滑坡栅格单元,确保所选非滑坡栅格单元是高概率的"非滑坡";最后采用SVM模型基于已知滑坡、所选非滑坡和环境因子对滑坡易发性进行评价。将提出的SOM-SVM模型用于三峡库区万州区滑坡易发性评价,并将得到的易发性结果与随机选取非滑坡的单独SVM模型结果做对比。结果显示SOM-SVM模型具有比单独SVM模型更高的成功率和预测率,表明SOM神经网络能更准确地选取非滑坡栅格单元。The non-landslide grid cells are selected randomly and/or subjectively when the machine learning models, such as the support vector machine （SVM）, are used to calculate the susceptibility indexes of regional landslides. However, it is difficult to determine whether the randomly selected non-landslide grid cells are reasonable ＂non-landslide＂ with very low susceptibility. To overcome this drawback, a model based on the combined clustering analysis and SVM is proposed. Firstly, the neural network with self-organizing mapping （SOM） is proposed to automatically classify the landslide susceptibility of all the grid cells into five classes- very low, low, moderate, high and very high susceptibility. Then, the reasonable non-landslide grid cells are selected from the area of very low susceptibility. Finally, the SVM is used to calculate the indexes of landslide susceptibility based on the recorded landslide grid cells, the selected non-landslide grid cells and the environmental factors. The proposed SOM-SVM model is used to calculate the susceptibility indexes of landslide in Wanzhou district of Three Gorges Reservoir area. The calculated results with the SOM-SVM model are compared with the results from the single SVM model which selects the non-landslide grid cells randomly. The results show that the SOM-SVM model has higher success and prediction rates than the single SVM. It is thus concluded that the non-landslide grid cells selected by the SOM neural network are more reasonable than the non-landslide grid cells selected randomly.

关 键 词：边坡工程 滑坡易发性 非滑坡栅格单元 自组织映射神经网络 支持向量机 

分 类 号：P642[天文地球—工程地质学]