摘要
针对近年来极端强降雨天气对深圳市带来的经济社会损失,提出一种改进的多层长短期记忆神经网络(long short-term memory, LSTM)道路积水大数据预测模型。以深圳市河湾流域58个内涝点的降雨积水数据为模型输入,并选取利用粒子群优化算法进行超参数寻优的回声状态网络(particle swarm optimization-echo state network, PSO-ESN)和传统LSTM为对比模型,比较三者在测试集上的纳什效率系数(Nash-Sutcliffe efficiency coefficient,E_(NS))。结果表明:改进的多层LSTM在26个有效点位中的21个点位上的测试集E_(NS)>0.8,达标率高达80.8%,远高于PSO-ESN和传统LSTM,证明该模型在研究区域进行道路积水预测工作的可行性。
In view of the huge social and economic property losses caused by extreme heavy rainfall in recent years in Shenzhen,this paper proposes a big data-driven road waterlogging prediction model based on improved multi-layer LSTM(long shortterm memory)deep learning network.Using the rainfall and waterlogging data of 58 waterlogging points in the Hewan basin of Shenzhen,and selecting a PSO-ESN(echo state network with hyperparameter optimization using particle swarm o9ptimization algorithm)and a traditional LSTM as comparison models,compare the E_(NS)(Nash-Sutcliffe Efficiency Coefficient)of these three models on the test set.The results show that the E_(NS) of the proposed model reach over 0.8 on test set at 21 of the 26 effective points,and the compliance rate is as high as 80.8%,which is much higher than those of PSO-ESN and traditional LSTM.,proving the feasibility of implementing road waterlogging prediction in study area with the proposed model.
作者
周涛
李成林
殷峻暹
金惠英
姚文才
黄磊
ZHOU Tao;LI Chenglin;YIN Junxian;JIN Huiying;YAO Wencai;HUANG Lei(Water Conservancy and Civil Engineering College,Xizang Agricultural and Animal Husbandry College,Linzhi 860000,China;Xizang Civil Engineering Water Conservancy and Electric Power Engineering Technology Research Center,Linzhi 860000,China;China Institute of Water Resources and Hydropower Research,Beijing 100038,China;Nanjing Jinma Intelligent Technology Co.,Ltd.,Nanjing 210000,China)
出处
《南水北调与水利科技(中英文)》
北大核心
2025年第S1期66-71,共6页
South-to-North Water Transfers and Water Science & Technology
作者简介
周涛(1999-),男,江苏溧阳人,主要从事水文学及水资源研究。E-mail:19851782600@163.com;通信作者:殷峻暹(1974-),男,正高级工程师,博士,主要从事水文学及水资源相关研究。E-mail:841829435@qq.com。
