基于长短期记忆网络的城市积水内涝风险预警  

Urban Waterlogging Risk Warning Based on Long Short-Term Memory Network

作　　者：许绘香 刘炜 XU Hui-xiang;LIU Wei(School of Information Engineering,Zhengzhou University of Technology,Zhengzhou Henan 450044,China;School of Software and Applied Science and Technology,Zhengzhou University,Zhengzhou Henan 450000,China)

机构地区：[1]郑州工程技术学院信息工程学院,河南郑州450044 [2]郑州大学软件与应用科技学院,河南郑州450000

出　　处：《计算机仿真》2024年第9期505-509,共5页Computer Simulation

基　　金：河南省高等学校重点科研项目(23B520024);河南省科技攻关项目(232102320015)。

摘　　要：城市积水内涝风险预警需要综合考虑多个因素的影响,如内涝空间与时间特征等,为增强城市积水内涝预警能力,及时应对潜在暴雨威胁,提出一种基于长短期记忆网络的城市积水内涝风险预警方法。利用内涝点核密度、不透水面密度指标明确积水蔓延面积,通过Moran指数得到积水空间分布汇聚特征,创建水文数据集和降雨数据集,把内涝时间特征提取问题转换为有监督学习问题,归一化处理后,将特征变量输入长短期记忆网络融合处理,明确内涝整体演变趋势;利用卷积神经网络预测未来一段时间的内涝水深,根据预测结果划分积水内涝风险等级,基于多源特征数据融合结果,使用单因子风险评估法与一维圣维南方程组完成内涝风险预警。仿真结果证实:所提方法可精准提取积水内涝特征,预警结果准确率高、效率快,可妥善处理极端暴雨威胁下积水内涝预警工作。The urban waterlogging risk early warning needs to comprehensively consider multiple factors,such as the spatial and temporal characteristics of waterlogging.In order to enhance the early warning level for urban waterlogging risk and respond to the potential rainstorm threat in time,this paper presented an early warning method for urban waterlogging risk based on long short-term memory network.Firstly,we used the core density of waterlogging points and the density of impervious surface to determine the spread area of waterlogging,and thus to obtain the spatial distribution and convergence characteristics of waterlogging according to Moran index.Secondly,we created hydrological data sets and rainfall data sets,and then transformed the problem of waterlogging time feature extraction into a problem of supervised learning.After normalization,we input the characteristic variables into the long-term and short-term memory network for fusion,thus determining the evolution trend of waterlogging.Thirdly,we used convolutional neural networks to predict the waterlogging depth in the future and divide the waterlogging risk level.Based on the fusion results of multi-source feature data,we used the single factor risk assessment method and one-dimensional Saint-Venant equations to complete the early warning of waterlogging risk.The simulation results show that the proposed method can accurately extract waterlogging characteristics,with high accuracy and efficiency,which can properly handle the early warning problem of waterlogging under the threat of extreme rainstorm.

关 键 词：多源信息融合 城市积水 内涝灾害 风险预警 时空特征提取 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]