基于ConvLSTM的时空域涌浪预报研究  

Research on spatiotemporal swell forecasting based on ConvLSTM

作　　者：吴昱翀 陶爱峰[1,2] 吕韬 曹力玮 王岗[1,2] WU Yu-chong;TAO Ai-feng;LYU Tao;CAO Li-wei;WANG Gang(College of Harbour,Costal and Offshore Engineering,Hohai University,Nanjing,Jiangsu 210098,China;Key Laboratory of Ministry of Education for Coastal Disaster and Protection,Hohai University,Nanjing,Jiangsu 210024,China)

机构地区：[1]河海大学港口海岸与近海工程学院,江苏南京210098 [2]河海大学海岸灾害及防护教育部重点实验室,江苏南京210024

出　　处：《中国港湾建设》2024年第12期1-9,共9页China Harbour Engineering

基　　金：国家重点研发计划资助(2023YFC3007900);国家自然科学基金项目(52271271);水利部重大科技项目(SKS-2022025)。

摘　　要：因风减弱、停止或转向后留在海上或来自其他海域的浪为涌浪,涌浪有着比风浪更低的波级,但是周期较大,易与船舶等浮式结构物发生共振,影响正常运行。为了避免涌浪造成的灾害,需要能够高效、精确预报涌浪的方法。基于Convolutional LSTM(ConvLSTM)模型搭建了神经网络,将涌浪的波高分布看作二维图像进行处理。网络将卷积运算的图像特征捕捉能力和长短期记忆神经网络(LSTM)模型的时序预测能力相结合,在网络学习过程中,既考虑了涌浪在空间上传播的特性,又考虑了时间上的变化特征。使用ERA5再分析数据集对网络进行训练,对东海(21°N-34°N,114°E-131°E)范围内的涌浪有效波高进行预测,预测结果与数据集较为吻合,最大相关性可达0.997。同时,与未考虑涌浪空间传播特性的模型对比预测效果有所提升。文中使用的方法为涌浪的预报研究提供了新的思路。Waves that remain at sea or come from other sea areas after the wind weakens,stops,or turns are called swells.Swell has a lower wave level than the wind wave,but the period is larger.Swell is easy to resonate with floating structures such as ships,which affects the normal operation.In order to avoid the damage caused by swells,it is necessary to have a method that can predict swells efficiently and accurately.A neural network was built based on the Convolutional LSTM(ConvLSTM)model to process the wave height distribution of swells was a two-dimensional image.The network combines the image feature capture ability of convolution operation and the time series prediction ability of Long Short Term Memory(LSTM)model,and considers both the spatial propagation characteristics of swell and the temporal variation characteristics in the network learning process.The ERA5 reanalysis dataset was used to train the network,and the significant wave height of the swell in the East China Sea(21°N—34°N,114°E—131°E)was predicted,and the prediction results were in good agreement with the dataset,with a maximum correlation of 0.997.At the same time,compared with the model that does not consider the spatial propagation characteristics of swells,the prediction effect is improved.The method used in this paper provides a new idea for the research on swell forecasting.

关 键 词：涌浪 卷积神经网络(CNN) LSTM ConvLSTM 神经网络 

分 类 号：U652.3[交通运输工程—港口、海岸及近海工程]