划分聚类法在短历时雨型设计中的应用  

Application of Partitioning Clustering in Short‑duration Storm Pattern Design

作　　者：金灿 袁文秀 周宏[1] 刘俊[1] JIN Can;YUAN Wen‑xiu;ZHOU Hong;LIU Jun(College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China;Jiangsu Provincial Water Resources Engineering Planning Office,Nanjing 210029,China)

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]江苏省水利工程规划办公室,江苏南京210029

出　　处：《中国给水排水》2024年第3期113-119,共7页China Water & Wastewater

基　　金：国家自然科学基金资助项目(41471015);国家重点研发计划项目(2018YFC0407201);中英联合科学创新基金资助项目(UUFRIP_100051)。

摘　　要：为更好地反映短历时暴雨特征,对雨型进行分类,基于四川省乐山国家基准气候站1981年—2021年逐分钟降雨数据资料,使用多种方法对90场历时60 min、78场历时120 min和86场历时180 min降雨选取最佳聚类数,采用K-means聚类和PAM聚类两种方法对雨型进行分析。结果表明,K-means聚类法对城市短历时暴雨雨型分类的结果比PAM聚类法更直观有效;乐山气候站历时60 min的雨型分为3类、历时120 min的雨型分为2类、历时180 min的雨型分为3类,单峰靠前的雨型最为常见。研究结果可为乐山市海绵城市建设、排水防涝规划等提供参考,同时可为机器学习在设计暴雨雨型研究中的应用提供新思路。In order to better reflect the characteristics of short‑duration rainstorms and classify the storm patterns,based on the minute‑by‑minute rainfall data from 1981 to 2021 at the National Benchmark Climate Station in Leshan,Sichuan Province,the best clustering numbers were selected for 90 rainstorms with 60 min duration,78 rainstorms with 120 min duration and 86 rainstorms with 180 min duration using various methods.K‑means and PAM partitioning clustering methods were used to analyze the storm pattern.The results showed that K‑means clustering was more intuitive and effective than PAM clustering in classifying urban short‑duration storm patterns.The storm patterns of Leshan Climate Station with duration of 60 min,120 min and 180 min were divided into three categories,two categories and three categories respectively,and the storm patterns with a single peak in front were the most common.The results are intended to provide references for sponge city construction,drainage and waterlogging control planning of Leshan City,and provide new ideas for the application of machine learning in the research of design storm patterns.

关 键 词：设计暴雨雨型 划分聚类 K-MEANS聚类 PAM聚类 乐山市 

分 类 号：TU992[建筑科学—市政工程]