基于机器学习短历时暴雨时空分布规律研究

城市内涝风险的精细化管理和防洪排涝市政工程的科学设计,需要对当地降雨的时空分布特征有深入的了解。而传统以单站雨型代表整个区域降雨特征的分析方法,不能满足这一要求。本文尝试将机器算法引入到暴雨时空分布特征研究中,以北京城区2004—2016年降雨资料为研究样本,利用动态聚类算法,提取北京城区短历时暴雨时空分布的动态特征。经分析,北京汛期的短历时暴雨时空分布特征,可以分为3种类型:(1)降雨自西北部山区移动到城中心区,逐渐扩散到城区;(2)降雨集中在城区西南部地区,逐渐向北部和城中心区扩散;(3)降雨集中在城区中心区和东部地区,基本不发生移动。研究结果表明,基于机器学习算法提取的暴雨时空分布特征,与实际暴雨时空动态发展趋势相符,并且有各自对应的降雨形成的不同物理机制,可为城区降雨设计、城市内涝风险管理等工作提供借鉴与参考。

An application of machine learning on examining spatial and temporal distribution of short duration rainstorm

It is necessary and important to understand the spatial and temporal distribution characteristics of rainfall in order to improve management of urban waterlogging risk and scientific design of municipal works for flood control and drainage. However, it is hard to derive accurate prediction based on monitoring data of a single station or the average of multiple stations. Availing of machine learning algorithm,this pa-per contributes to investigating rainfall spatial and temporal distribution characteristics. Specifically,the tech-nique of dynamic clustering algorithm examines the spatial and temporal distribution characteristics of rain-fall in Beijing urban area during recent 10 years. The result shows that there are three types of rainstorms in Beijing urban area. Rainstorms of type 1 move from the northwest to the center of Beijing,then spreads to the eastern part of the urban area; rainstorms of type 2 occurs in the southwestern region of the urban area, and gradually northward, but there is no rainfall in the mountainous northwest; rainstorms of type 3 are concentrated in the central and eastern regions,and basically does not move. The results are consistent with the actual rainstorm process. It provides references for urban rainfall design and urban waterlogging risk management.