基于大数据的极端暴雨事件下城市道路交通及人群活动时空响应

随着全球气候变化加剧,极端降雨增多,暴雨内涝灾害频发,严重威胁城市的可持续发展。快速掌握暴雨给城市交通及人群的影响,有助于提高灾害应急管理水平和事件响应能力。利用实时动态的交通路况信息和手机定位请求数据,通过一种融合STL时序分解技术与极端学生化偏差统计检验的时间序列异常探测方法,监测和分析暴雨内涝灾害事件中,城市道路交通和人群活动的时空响应特征,并以2018年7月16日发生在北京的极端暴雨事件为例开展实证研究。研究结果显示,在降雨集中的早、晚高峰两个时段（8—9时、18—19时）,市区的拥堵道路数量超出往常水平最高可达150%,异常检测分析显示拥堵道路数量和交通拥堵指数均达到异常甚至极端水平。人群活动的异常响应分析结果显示,暴雨事件引起定位请求量异常升高、异常点增多,且异常点的空间分布与1 h前的降雨量分布存在较高相关性。以上结果不仅证明了大数据及异常检测方法对于快速洞察暴雨事件对城市交通及人群影响的有效性,也为城市暴雨内涝灾害的应急响应与管理提供了新的技术手段。

Spatio-temporal responses of urban road traffic and human activities in an extreme rainfall event using big data

As global climate change intensifies, extreme rainfalls and floods become more frequent and pose a serious threat to urban sustainable development. Fast assessment of the rainfall disaster impact upon urban traffic and population plays an important role in improving disaster emergency management and incident response capabilities. This study adopts a time series anomaly detection method to discover and quantify the impact of rainfall-triggered flood on road traffic and human activities using real-time traffic condition information and mobile phone location request data. The anomaly detection method combines the STL time series decomposition technique and the extreme student deviation statistics to identify the response characteristics of traffic data and location requests during the event. The extreme rainfall event that occurred in Beijing on July 16, 2018 is used as a case study to examine the method effectiveness. The results show that the precipitation peaked in the morning and evening rush hours, during which the number of congested roads exceeded the average level by up to 150%.The anomaly detection analysis indicates that the number of congested roads and the traffic congestion index reached the outlier level. The anomaly analysis of human activity responses shows that the heavy rainfall event also caused an abnormal increase in the number of location requests, and the spatial distribution of the anomalous grids was highly correlated with the rainfall distribution one hour before. The above results not only prove the effectiveness of the big data and the anomaly detection method in understanding the impact of heavy rainfall events on urban traffic and population, but also provide new means for urban emergency response and management against rainfall disasters.