降雨型滑坡灾害的特征聚合决策树预测模型  

Feature Aggregation Decision Tree Prediction Model for Rainfall Landslide Disaster

作　　者：赵晓萌[1] 卫星君[2] 王娜[1] 雷向杰[1] ZHAO Xiaomeng;WEI Xingjun;WANG Na;LEI Xiangjie(Shaanxi Provincial Climate Center,Xi’an 710014,China;School of Electrical and Information Engineering,Shaanxi Energy Institute,Xianyang 712000,China)

机构地区：[1]陕西省气候中心,陕西西安710014 [2]陕西能源职业技术学院,陕西咸阳712000

出　　处：《灾害学》2020年第1期27-31,共5页Journal of Catastrophology

基　　金：国家重点基础研究发展计划(973)(2013CB430202);中国气象局气候变化专项项目(CCSF201845);陕西省气象局秦岭和黄土高原生态环境气象重点实验室青年基金课题(2019Y-7)

摘　　要：为了有效预警降雨型滑坡灾害,以秦巴山区为研究区域,采集并处理大量不同时段降雨数据,构成降雨特征属性。利用Fisher最优分割,对降雨特征属性值进行分段统计,提出特征聚合转换表。使用信息增益和预测反馈筛选影响滑坡灾害的有效降雨特征,为预测模型提供有效数据集。利用特征聚合转换表和有效降雨特征,改进决策树,构建特征聚合决策树预测模型,进而提高预测效率和预测准确率。分析决策树的深度和叶子节点个数,给出决策树的反馈执行度,表明使用特征聚合转换表的决策树更优。比较特征聚合决策树、决策树、朴素贝叶斯和逻辑回归预测模型,结果表明,特征聚合决策树预测模型对降雨型滑坡灾害有更高的预测准确率,且平均预测准确率较高。In order to forewarn the rain-type landslide disaster effectively, the Qinba Mountain area is used as the research area. In the research, a large number of rainfall data are collected and processed to form rainfall feature attributes. Fisher optimal segmentation is used to propose a feature aggregation conversion table by segmentation statistics of rainfall feature attribute values. Effective rainfall characteristics of landslide hazards are screened by information gain and predictive feedback to provide valid data sets for predictive model. Feature aggregation conversion table and effective rainfall characteristics are used to improve the decision tree. Meanwhile, the prediction model of the feature aggregation decision tree is constructed to achieve high prediction efficiency and accuracy. Having analyzed the depth of the decision tree and the node number of the leaves, the feedback execution of the decision tree shows that the decision tree of the feature aggregation conversion table is better. Feature aggregation decision trees, decision trees, naive Bayesian prediction models and logistic regression are compared. The results show that the prediction model the feature aggregation decision tree has higher prediction accuracy for rainfall landslide disasters, and the average prediction accuracy is higher.

关 键 词：降雨 滑坡灾害 信息增益 最优分割 决策树 

分 类 号：X43[环境科学与工程—灾害防治] X915.5