基于机器学习快速预报模型的城市洪涝预报预警系统研究及应用  

Study and Application of Urban Flood Forecasting and Early Warning System basedon Machine Learning Rapid Forecasting Model

作　　者：史超 王兴桦 吴新垒 黄玉 李智星 毛叶丰 SHI Chao;WANG Xinghua;WU Xinlei;Huang Yu;LI Zhixing;MAO Yefeng(Ningbo Haishu District Flood Control and Drought Relief Center,Ningbo 315000,China;State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China,Xi'an University of Technology,Xi'an 710048,China;Ningbo Big Data Investment Development Co.,Ltd.,Ningbo 315100,China;Sichuan Shuifa Survey,Design and Research Co.,Ltd.,Chengdu 610000,China;Zhongzhi Shuike(Ningbo)Technology Co.,Ltd,Ningbo 315100,China)

机构地区：[1]宁波市海曙区水旱灾害防御中心,浙江宁波315000 [2]西安理工大学省部共建西北旱区生态水利工程国家重点实验室,西安710048 [3]宁波市大数据投资发展有限公司,浙江宁波315100 [4]四川水发勘测设计研究有限公司,成都610000 [5]中智水科(宁波)科技有限公司,浙江宁波315100

出　　处：《西北水电》2023年第2期12-19,共8页Northwest Hydropower

基　　金：国家自然科学基金面上项目(52079106)

摘　　要：为弥补历史降雨积涝数据的不足,实现对暴雨导致的城区洪涝全过程的精准预测,采用城市洪涝水文水动力模型模拟暴雨-内涝过程,设计各种降雨情景获取全面的降雨积涝数据,运用用机器学习算法建立降雨-积涝关系,构建了一个城市洪涝预报决策系统,并以宁波海曙城区为例进行实证,结果表明:采用的KNN机器学习算法可有效驱动降雨-内涝数据库,准确表征降雨径流过程,对城市内涝情况进行预报预警;将宁波海曙城区城市洪涝情况在系统上进行城市洪涝预报和复盘结果对比表明,降雨不确定性是导致模型预报结果偏差的一个主要原因,应获取更准确的短临降雨数据;地形在生成过程中存在人为修正情况,且城市建设对局部地形有一定影响,需对局部地形异常区域进行实测并修正,根据真实积水数据优化模型。该城市洪涝预报决策系统可实现城市洪涝快速地准确预报,为城市洪涝预警和应急管理提供支撑。In order to make up for the shortage of historical rainfall waterlogging data and realize the accurate prediction of the whole process of urban flood caused by rainstorm,urban flood hydrodynamic model is used to simulate the process of rainstorm-waterlogging,various rainfall scenarios are designed to obtain comprehensive rainfall waterlogging data,machine learning algorithm is used to establish the relationship between rainfall and waterlogging,and an urban flood forecasting decision-making system is constructed which is demonstrated with the example of Ningbo Haishu District.The results show that the KNN machine learning algorithm can effectively drive the rainfall-waterlogging database,accurately characterize the rainfall-runoff process,and forecast and warn the urban waterlogging situation;The comparison of urban flood forecasting and re-assessment results based on the platform for the urban flood situation in urban area of Ningbo Haishu District shows that rainfall uncertainty is a major reason for the deviation of the model forecasting results,and more accurate short-term and imminent rainfall data should be obtained;In the process of terrain generation, there are artificial corrections, and urban construction has a certain impact on local terrain. Therefore, it is necessary to survey and correct local terrain abnormal areas, and optimize the model based on actual waterlogging data.The urban flood forecasting and decision-making system can achieve rapid and accurate urban flood forecasting, providing support for urban flood early warning and emergency management.

关 键 词：城市洪涝 GAST水动力数值模型 机器学习算法 预报预警 

分 类 号：TV124[水利工程—水文学及水资源]