基于SOM-I-SVM耦合模型的滑坡易发性评价

Landslide susceptibility assessment based on the SOM-I-SVM model

在使用机器学习模型对滑坡进行易发性评价时,通常会在滑坡影响范围之外随机选取非滑坡样本点,具有一定的误差。为了提高滑坡易发性评价的精度,将自组织映射(self-organizing map,SOM)神经网络、信息量模型(information,I)以及支持向量机模型(support vector machine,SVM)进行耦合,提出一种基于SOM-I-SVM模型的滑坡易发性评价方法,并将SOM神经网络与K均值聚类算法进行对比,验证模型的可靠性。以十堰市茅箭区为例,首先通过对环境因子的相关性及重要性分析,筛选出距水系距离、坡度、降雨量、距构造距离、相对高差、距道路距离、地层岩性等7个因子,建立滑坡易发性评价指标体系,在此基础上计算出各因子的分级信息量值,并作为模型的输入变量进行滑坡易发性评价。分别采用SOM神经网络和K均值聚类算法选取非滑坡样本,然后将样本数据集代入I-SVM模型预测滑坡易发性。将SVM、I-SVM、KMeans-I-SVM、SOM-I-SVM等4种模型预测精度进行对比,其ROC曲线下面积(AUC)分别为0.82,0.88,0.90,0.91,说明SOM-I-SVM模型能有效提高滑坡易发性预测准确率。

When using machine learning models for landslide susceptibility evaluation,the non-landslide sample points are usually selected randomly outside the landslide influence area,leading to a certain error.To improve the accuracy of landslide susceptibility evaluation,this paper couples the self-organizing map(SOM)neural network,information(I)model,and support vector machine(SVM)model,and proposes a SOM-I-SVM model-based method of landslide susceptibility evaluation,comparing with K-means clustering to verify the reliability of this model.The Maojian District of the city of Shiyan is taken as an example,and seven factors of the distance from water system,slope,rainfall,distance from structure,relative height difference,distance from road,stratigraphic lithology are selected by correlation and importance analyses of environmental factors to establish a landslide susceptibility evaluation system.Based on these,the graded information values of each factor are calculated and used as input variables for landslide susceptibility evaluation.The SOM neural network and K-means clustering are used to select non-landslide samples,and the sample data set is substituted into the I-SVM model to predict landslide susceptibility.The prediction accuracies of the four models,SVM,I-SVM,KMeans-I-SVM and SOM-I-SVM,are compared,and the area under the ROC curve(AUC values)are 0.82,0.88,0.90 and 0.91,indicating that the SOM-I-SVM model can effectively improve the accuracy of landslide susceptibility prediction.