基于神经网络模型对风暴潮特征分析及预测——以深圳赤湾站和南澳站为例  

Characteristics analysis and tide level prediction of storm surge based on neural network model——Taking Chiwan and Nan'ao in Shenzhen City as an example

作　　者：刘媛媛[1,2] 刘业森[1,2] 张丽[3] 力梅 穆杰[1,2] LIU Yuanyuan;LIU Yesen;ZHANG Li;LI Mei;MU Jie(China Institute of Water Resources and Hydropower Research,Beijing 100038;Research Center on Flood and Drought Disaster Reduction of the Ministry of Water Resources,Beijing 100038;Shenzhen National Climate Observatory,Shenzhen 518040)

机构地区：[1]中国水利水电科学研究院,北京100038 [2]水利部防洪抗旱减灾工程技术研究中心,北京100038 [3]深圳市国家气候观象台,深圳519082

出　　处：《中国防汛抗旱》2022年第7期66-71,共6页China Flood & Drought Management

基　　金：国家自然科学基金项目(52009147)。

摘　　要：我国沿海城市经常遭受台风的影响,台风带来的强降雨如果遭遇风暴潮高潮位,会造成严重的洪涝灾害。在台风到来之前,对风暴潮位进行科学准确的分析和预测,及时合理地调度工程,对降低沿海城市的洪涝灾害风险具有重要的意义。在分析了风暴潮增水影响因素的基础上,在赤湾站和南澳站分别建立了长短期记忆人工神经网络预测模型。结果表明,处于不同位置的潮位站,风暴潮位影响因素及前时序n各不相同,利用前3 h潮位和风速预测南澳站的风暴潮位,误差最小,而赤湾站则是利用前6 h的潮位和风速预测,误差最小。Coastal cities of China are often affected by typhoons.Extreme rainfall brought by typhoons will cause serious flood and waterlogging if it encounters the high tide of storm surge.Before the arrival of typhoon,scientific and accurate analysis and prediction of storm tide level and timely and reasonable scheduling of projects are of great significance to reduce the risk of waterlogging disasters in coastal cities.Based on the analysis of the influencing factors of storm surges,the LSTM(Long-Short Term Memory)artificial neural network prediction models of Chiwan station and Nan'ao station are established.The results show that the influencing factors and pre time sequence n of storm tide level are different at different tide level stations.The error is the smallest by using the tide level and wind speed in the first 3 hours to predict the storm tide level at Nan'ao station,and predicted with the minimum error by using the tide level and wind speed in the first 6 hours at Chiwan station.

关 键 词：风暴潮 LSTM神经网络 智能预测 深圳 

分 类 号：P731.23[天文地球—海洋科学]