基于CF融入SSA优化SVM和RF模型的滑坡易发性评价

Landslide susceptibility evaluation based on CF integrated with SSA to optimize SVM and RF models

针对传统的区域滑坡易发性评价建模过程可能存在的样本数据量纲不统一以及模型参数选取误差等问题,本文以陕西省留坝县为研究区,选取高程、坡度、水系、降雨量、地层岩性等10个评价因子,采用确定性系数模型(CF)计算各评价因子的敏感值作为支持向量机模型(SVM)和随机森林模型(RF)的输入样本属性值,引入麻雀搜索算法(SSA)分别对SVM模型和RF模型的参数进行优化,获取最优参数对两种模型进行训练,最终构建了CF-SSA-SVM和CF-SSA-RF模型,从而对整个研究区进行预测,完成滑坡易发性评价,并通过受试者工作特征曲线(ROC)对两种模型进行精度验证。结果表明,两种模型的评价结果均有较多滑坡点落在极高易发区,无滑坡点落在极低易发区,评价结果均有较高的准确率。其中,CF-SSA-RF模型的成功率和预测率曲线AUC值分别为0.994和0.940,高于CF-SSA-SVM模型;并以三处典型滑坡为例进行验证,结果显示易发性分区与历史滑坡点分布较为吻合。进一步表明CF-SSA-RF模型更适用于留坝县的滑坡易发性评价,为当地滑坡灾害风险评估提供了指导依据。

For the traditional modeling process of intra-regional landslide susceptibility evaluation,there may be problems such as non-uniformity of sample data outline and errors in the selection of model parameters.This paper takes Liuba County of Shaanxi Province as the research area,selects 10 evaluation factors such as elevation,slope,water system,rainfall,stratigraphic lithology,etc.,and uses the certainty factor model(CF)to calculate the sensitivity of each evaluation factor as a support vector machine model(SVM)and random forest model(RF)input sample attribute values;it introduces the sparrow search algorithm(SSA)to optimize the parameters of SVM model and RF model respectively,obtains the optimal parameters to train the two models,and finally constructs CF-SSA-SVM and CF-SSA-RF models,which can predict the entire study area,complete the landslide susceptibility evaluation,and verify the accuracy of the two models through the receiver operating characteristic curve(ROC).The results show that the evaluation results by the two models have more landslide points in the extremely high-prone areas,and no landslide points in the extremely low-prone areas,and that the evaluation results are of high accuracy.Among them,the AUC values at the success rate and prediction rate curves of the CF-SSA-RF model are 0.994 and 0.940,respectively,which are higher than those by the CF-SSA-SVM model;verified by three typical landslides,the results show that the prone zones and historical landslide points are relatively consistent.It further shows that the CF-SSA-RF model is more suitable for the landslide susceptibility evaluation research in Liuba County,providing a guiding basis for the local landslide disaster risk assessment.