徐胜华, 刘纪平, 王想红, 张玉, 林荣福, 张蒙, 刘猛猛, 姜涛. 熵指数融入支持向量机的滑坡灾害易发性评价方法—以陕西省为例[J]. 武汉大学学报 ( 信息科学版), 2020, 45(8): 1214-1222. DOI: 10.13203/j.whugis20200109
引用本文: 徐胜华, 刘纪平, 王想红, 张玉, 林荣福, 张蒙, 刘猛猛, 姜涛. 熵指数融入支持向量机的滑坡灾害易发性评价方法—以陕西省为例[J]. 武汉大学学报 ( 信息科学版), 2020, 45(8): 1214-1222. DOI: 10.13203/j.whugis20200109
XU Shenghua, LIU Jiping, WANG Xianghong, ZHANG Yu, LIN Rongfu, ZHANG Meng, LIU Mengmeng, JIANG Tao. Landslide Susceptibility Assessment Method Incorporating Index of Entropy Based on Support Vector Machine: A Case Study of Shaanxi Province[J]. Geomatics and Information Science of Wuhan University, 2020, 45(8): 1214-1222. DOI: 10.13203/j.whugis20200109
Citation: XU Shenghua, LIU Jiping, WANG Xianghong, ZHANG Yu, LIN Rongfu, ZHANG Meng, LIU Mengmeng, JIANG Tao. Landslide Susceptibility Assessment Method Incorporating Index of Entropy Based on Support Vector Machine: A Case Study of Shaanxi Province[J]. Geomatics and Information Science of Wuhan University, 2020, 45(8): 1214-1222. DOI: 10.13203/j.whugis20200109

熵指数融入支持向量机的滑坡灾害易发性评价方法—以陕西省为例

Landslide Susceptibility Assessment Method Incorporating Index of Entropy Based on Support Vector Machine: A Case Study of Shaanxi Province

  • 摘要: 滑坡灾害易发性评价可为滑坡灾害风险管理、国土空间规划及滑坡监测提供科学依据。针对现有滑坡灾害易发性评价模型无法消除易发性评价指标因子在量纲、性质等方面的差异,尚未考虑易发性评价指标因子与滑坡灾害相关性,以及精度较高的经典机器学习模型训练效率较低、参数选取困难等问题,引入熵指数(index of entropy,IOE)和粒子群优化(particle swarm optimization,PSO)算法,提出IOE融入支持向量机(support vector machine, SVM)的滑坡灾害易发性评价方法。首先,基于滑坡灾害易发性评价指标因子,利用IOE模型计算SVM的调节因子;然后,采用PSO算法迭代求解SVM最优解,根据SVM二分类得到的隶属度来区分滑坡灾害易发性;最后,以陕西省作为实验区,从滑坡灾害易发性分区图、分区统计及评价模型精度3个方面将所提方法与SVM方法进行了对比,实验结果表明所提方法的准确性、可靠性优于SVM方法。

     

    Abstract:
      Objectives  The landslide susceptibility assessment can provide important scientific basis for landslide hazard risk management, land space planning and landslide monitoring. Shaanxi Province, located in northwestern China, is a key area for the prevention and control of landslides in the country. Shaanxi Province has active geological tectonic movements,complex geological conditions,and is affected by various factors such as engineering construction, land resource utilization, mineral resource development, and heavy rainfall. The effects of landslide hazards have increased significantly. Therefore, it is of great practical significance to monitor and predict landslide disasters in Shaanxi Province.
      Methods   Considering that the existing landslide hazard susceptibility evaluation models cannot eliminate the differences in dimensionality and properties of susceptibility evaluation index factors, we take into account the correlation between susceptibility evaluation index factors and landslide hazard, especially the methods based on classic machine learning with high accuracy still exist some problems such as low training efficiency and difficult parameter selection, landslide susceptibility assessment method incorporating index of entropy based on support vector machine (SVM) is presented using index of entropy (IOE) and particle swarm optimization algorithm(PSO).First,based on the landslide susceptibility evaluation index factors,8 evaluation factors are selected, such as elevation, slope, aspect, distance from residential area, distance from road, distance from river, lithology and cumulative rainfall. In order to ensure the balance of sample units and the uniform distribution and reduction of non‑landslide hazard unit error,non‑hazard units with the same number of hazard units are selected randomly in the area 1 km away from the existing hazard point. All the non‑hazard units and hazard units are taken as the training data set and verification data set. Then the IOE model is used to calculate the adjustment factor supporting SVM, the PSO algorithm applied to iteratively solve the optimal penalty factor and radial basis kernel function of SVM, and the susceptibility of landslide hazards is obtained according to the membership of the SVM binary classification. At last, the susceptibility partition is divided by the natural breakpoint method.
      Results   Selecting Shaanxi Province as the experimental area, the proposed method is compared with the SVM method from the three aspects of landslide hazard susceptibility partition map, partition statistics and evaluation model accuracy. Compared with the SVM model, the hazard susceptibility is improved one level is more reasonable for hazard susceptibility zoning in some areas with hazard point distribution. With the increase of hazard susceptibility, the percent of landslide hazard points in each subregion of the two models also gradually increased, reaching the highest in extremely high incidence areas, especially in extremely high incidence areas. The percent of landslide hazard points in the proposed method reached 58.64%, higher than the SVM method.
      Conclusions   In the landslide hazard‑prone area divided by the proposed method, the point density of hazard points is higher, which is more in line with the actual hazard distribution.The receiver operating characteristic(ROC) curve shows that the area under curve(AUC) value of the proposed method is 0.89, and the SVM method is 0.82, which is higher than SVM. Based on the experimental results,it is shown that the accuracy and reliability of the proposed method is superior to the SVM method.

     

/

返回文章
返回