基于PSO-SVM模型的滑坡易发性评价

滑坡易发性评价是区域滑坡预警和评估工作的前期准备。滑坡致灾因子的有效选取以及评价模型的构建成为当前滑坡预测研究中的难点问题。以府谷县作为研究区,借助多种技术手段将数字高程模型(DEM)、地质图、路网图、遥感影像图等多源数据进行融合,提取了地形地貌、地层岩性以及地表覆盖等滑坡孕灾环境因子和降雨量、人类工程活动等诱发因素的特征属性作为评价指标。在此基础上,对提取的各因子的相关性进行分析,剔除了地形起伏度因子。采用粒子群优化(PSO)算法对支持向量机(SVM)模型的参数进行优化,得到最佳参数组合为惩罚因子c=1.42,核参数σ=1.15,将最优参数组合代入支持向量机模型中,构建出粒子群优化算法-支持向量机模型(PSO-SVM),并将其用于研究区滑坡易发性定量评价中。最后分别采用ROC曲线与Kappa系数对PSO-SVM模型性能的优越性进行检验,结果表明,PSO-SVM模型成功率与预测率分别为0.931和0.917,训练集与测试集的预测精度分别为79.17%、76.67%。研究结果可以为从事滑坡预测评价工作者提供决策参考。

Assessment of Landslide Susceptibility Based on PSO-SVM Model

Landslide susceptibility assessment is a precondition of early warning and evaluation for regional landslide. Effective selection of hazard-inducing factors and establishment of assessment model are challenging in the prediction of landslide hazards. On the basis of the fusion of multi-source data including digital elevation model (DEM), geological map, road network map, and remote sensing image of Fugu County as a case study, the environmental factors such as landform and geomorphology, formation lithology and ground cover as well as inducing factors such as rainfall and human engineering activity were extracted as assessment indicators. The correlations among the extracted factors were analyzed and the topographic relief factor was eliminated. Furthermore, the particle swarm optimization (PSO) algorithm was adopted to optimize the parameters of support vector machine(SVM) model. The optimal parameters (penalty parameter c=1.42 and kernel parameter σ=1.15) were incorporated into the SVM model to establish the PSO-SVM model for landslide susceptibility assessment. The performance of the model was tested by the receiver operate curve (ROC) and Kappa coefficient, and results revealed that the success rate and the prediction rate of the PSO-SVM model were 0.931 and 0.917, respectively, and the prediction accuracy of train data and test data were 79.17% and 76.67%, respectively.