相关向量机在地震滑坡敏感性分析中的应用

 地震滑坡敏感性分析是地震次生灾害研究的重点内容之一。数据量大且致灾因素复杂是研究地震滑坡问题的难点。在对已有敏感性分析模型研究的基础上，以芦山地震为例，选取地面高程、坡度、坡向、地层、斜坡形态、斜坡结构、距断层平均距离、距水系平均距离、地震峰值加速度9个地震滑坡评价因子，建立基于遗传算法的相关向量机（GA-RVM）敏感性分析模型，生成地震滑坡敏感性区划图，统计结果显示滑坡正确率为99.74%，滑坡密度在极高敏感区达到27.4057个/km²。结果表明，相对于基于遗传算法的支持向量机，GA-RVM获得了更高的预测精度，可为进一步完成地震灾害预防提供依据。

Application of relevance vector machine to earthquake-induced landslide susceptibility assessment

The earthquake-induced landslide susceptibility assessment is one of the important parts in the researches of secondary disasters of earthquake. In view of the large amount of data, the rich information, the complex relationship, it is a very difficult task. This paper takes Lushan in the 2013 Lushan earthquake as the research area. Massive landslides were triggered by this earthquake. Among these landslides, 226 landslides are interpreted based on aerial photographs in Lushan, which are verified by the field investigation. Then 9 impact factors are selected by the Pearson correlation analysis, including the elevation, the slope, the aspect, the curvature classification, the slope structure, the lithology, the distance from drainages, the distance from faults, and the peak ground acceleration. The relevant vector machine(RVM) is a new learning procedure based on the statistical learning theory, and a genetic algorithm(GA) is adopted to optimize the parameter of the RVM. The proposed GA-RVM model is used to calculate the landslide susceptibility value, to produce susceptibility zoning. The statistical data of the susceptibility zoning are as follows:(1)the accuracy rate of the landslides is 99.74%; (2)the density of the landslides in a high susceptibility zoning is 27.4057 per square kilometers. The result shows that the relevant vector machine model is better than the support vector machine and is suitable for the earthquake-induced landslide susceptibility assessment and the earthquake disaster prevention.