首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于多变量LSTM神经网络模型的风暴潮临近预报
引用本文:刘媛媛,张丽,李磊,刘业森,陈柏纬,张文海.基于多变量LSTM神经网络模型的风暴潮临近预报[J].海洋通报,2020,39(6):689-694.
作者姓名:刘媛媛  张丽  李磊  刘业森  陈柏纬  张文海
作者单位:中国水利水电科学研究院,北京100038;深圳市国家气候观象台 广东深圳519082;中山大学大气科学学院,广东珠海519082;香港天文台,香港九龙999077;深圳市强风暴研究院,广东深圳518057
基金项目:国家重点研发计划 (2016YFC0803107;2016YFC0803109)
摘    要:台风的风暴潮是台风引发的一种重要次生灾害,对沿海城市带来的威胁是多方面的。及时准确地预报风暴潮,对沿海地区采取合理措施减少人员伤亡和经济损失具有重要意义。本文利用长短期记忆神经网络 (LSTM) 模型,综合考虑风速、 风向、气压等气象因素和前时序的潮位数据,建立了风暴潮的临近预报模型。结果表明,基于 LSTM 的临近预报模型具有相当的预报技巧,利用前时序的风速和风向数据以及潮位数据建立的模型可对风暴潮潮位进行准确地预测。研究还表明,仅考虑前时序潮位的预测模型误差最大,考虑气压后的模型预测能力有一定进步,而考虑风的要素以后,预测的效果提升更为明显。

关 键 词:LSTM神经网络模型  热带气旋  风暴潮  临近预报
收稿时间:2020/4/8 0:00:00
修稿时间:2020/5/7 0:00:00

Storm surge nowcasting based on multivariable LSTM neural network model
LIU Yuanyuan,ZHANG Li,LI Lei,LIU Yesen,CHAN Pakwai,ZHANG Wenhai.Storm surge nowcasting based on multivariable LSTM neural network model[J].Marine Science Bulletin,2020,39(6):689-694.
Authors:LIU Yuanyuan  ZHANG Li  LI Lei  LIU Yesen  CHAN Pakwai  ZHANG Wenhai
Institution:China Institute of Water Resources and Hydropower Research, Beijing 100038, China;Shenzhen National Climate Observatory,Shenzhen 518040, China;School of Atmospheric Sciences, Sun Yat-Sen University, Zhuhai, 519082, China;Hong Kong Observatory, Hong Kong 999077, China; Shenzhen Academy of Severe Storms Science, Shenzhen 518057, China
Abstract:A storm surge is an important secondary disaster caused by typhoons, and its threat to coastal cities is manifold. It is of great significance to forecast storm surges timely and accurately, and to adopt reasonable measures to reduce casualties and economic losses in coastal areas. This paper proposes a new algorithm which comprehensively considers wind speed, wind direction, atmospheric pressure, and other factors, as well as the tidal level observation data of preceding time series,and establishes a storm surge nowcasting model by using the long short-term memory (LSTM) neural network algorithm. The results reveal that the LSTM-based nowcasting model has considerable forecasting skills and that the forecasting error is related to the time level advance of the preceding data used. The study also reveals that the forecasting model which only considers the tidal level of the preceding time series has the largest error. Furthermore, the forecasting ability of the model improves to an extent after the atmospheric pressure is taken into account, while the forecasting effect improves more significantly after wind is considered.
Keywords:long short-term memory  tropical cyclone  storm surge  nowcasting
本文献已被 万方数据 等数据库收录!
点击此处可从《海洋通报》浏览原始摘要信息
点击此处可从《海洋通报》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号