基于微信文本的台风灾情快速评估方法研究

Rapid Assessment Method of Typhoon Disaster Based on WeChat Text

利用Python程序采集了与2019年台风“利奇马”和“白鹿”相关的1 036篇有效微信文章,抽取其中的灾情信息,基于词云图分析了灾害应急信息类型并给出台风“利奇马”中各省灾情大小的排序;构建了省级台风灾情指数并以此验证上述灾情排序结果.结果表明:(1)在台风的不同阶段,大众的关注内容反映出不同的灾害应急信息类型,在台风登陆前,主要为交通信息、天气预警;台风登陆时(后)主要为交通信息、天气预警、伤亡及营救、次生灾害、电力中断、事件/人物追踪、灾后恢复等;台风消亡后,主要为伤亡及营救、灾后恢复;基于微信文本数据推测的各省灾情排序为浙江、山东、上海、安徽、江苏、福建、中国台湾、湖南;(2)台风“利奇马”的灾情指数为11.57,处于大灾区间范围内,台风“白鹿”灾情指数为1.33,处于轻灾区间范围内;在省级台风灾情指数中,各省灾情排序依次为浙江、山东、安徽、江苏、上海、辽宁、河北、吉林、福建,与基于微信文本数据推测的各省灾情排序基本吻合.

In this thesis, we collected 1 036 valid WeChat articles related to Super Typhoon “Lekima” and Severe Tropical Storm “Bailu” in 2019 using Python program, extracted disaster information from them, analyzed the type of disaster emergency information based on the word cloud, ranked the disaster size of each province in Super Typhoon “Lekima”,and constructed a provincial typhoon disaster index to verify the above disaster ranking results.The results show that:(1)At different stages of the typhoon, the public’s attention reflects different types of disaster emergency information.Before the typhoon landfall, it is mainly traffic information and weather warning and at(after) the typhoon landfall, it is mainly traffic information, weather warning, casualties and rescue, secondary disasters, power interruption, event/personal tracking, and post-disaster recovery.After the typhoon dissipates, it is mainly casualties and rescue, and post-disaster recovery.The ranking of disaster situation in each province inferred from WeChat text data is Zhejiang, Shandong, Shanghai, Anhui, Jiangsu, Fujian, Taiwan, and Hunan.(2)The disaster index of Typhoon “Lekima” is 11.57,which is within the range of major disaster, and the disaster index of Severe Tropical Storm “Bailu” is 1.33,which is within the range of light disaster.In the provincial typhoon disaster index, the disaster ranking of each province is Zhejiang, Shandong, Anhui, Jiangsu, Shanghai, Liaoning, Hebei, Jilin and Fujian, which is basically consistent with the disaster ranking of each province based on WeChat text data.