基于SOCP-MKRVM的区域滑坡敏感性分析研究∗

摘要

In order to improve the prediction ability of relevance vector machine ( RVM) for the regional landslide suscepti-bility assessment, a prediction model of multiple-kernel RVM based on second-order cone programming ( SOCP-MKRVM) was proposed. Taking the low hilly area of Sichuan Province as example, eight landslide-predisposing factors were selected to train the RVM prediction model, and two methods which include the receiver-operating characteristic curve ( ROC) and the landslides dot density were used to verify the prediction results of the model. Through the contrastive analysis with the single kernel RVM model, the results showed that the SOCP-MKRVM model improved the assessment ability of the regional land-slide susceptibility. The prediction accuracy increased to 71. 33%, the area under the ROC curve reached 0. 741, and the distribution of landslide dot density was more reasonable, with the sum of two low susceptibility areas as 0. 89/100 km2 and the sum of two high susceptibility area as 6. 54/100 km2 .%为了提高相关向量机( RVM)在区域滑坡敏感性评价中的预测能力,提出了基于二阶锥规划的多核相关向量机( SOCP-MKRVM)预测模型。以四川省低山丘陵区为例,选取了8个滑坡孕灾因子训练RVM预测模型,并分别运用受试者工作特征曲线( ROC)和滑坡点密度2种方法对预测结果进行验证。通过与单核RVM模型的对比分析,结果表明：SOCP-MKRVM模型提高了对区域滑坡敏感性的评价能力,预测精度提高到71.33%,ROC曲线下面积达到0．741,滑坡点密度分布更加合理,两低敏感区之和为0．89个/100 km2,两高敏感区之和为6.54个/100 km2。