基于多区域协同的海洋波浪周期时空预测

打开文本图片集

中图分类号：P731.33；TP391 文献标识码：A

Abstract： Existing wave prediction methods （single-point， multi-point， single-region） often overlook spatial correlation and dynamic inter-regional connections， making it difficult to capture cross-regional dynamic changes. To address this issue， this paper proposed a Multi-Region Collaborative Spatio-Temporal Graph Convolutional Network （MRCSTGCN）. This model employs a temporal attention block to capture temporal dependencies，while a spatio-temporal feature extractor recalibrates global information to fuse multi-region multi-scale features. A Graph Convolutional Network （GCN） explicitly models spatio-temporal relationships between regions using prior knowledge； the Transformer module learns complex implicit spatio-temporal patterns，and a gated fusion module dynamically integrates the outputs of these two components. Experiments conducted on the wave period data from the fifth-generation reanalysis dataset of ECMWF demonstrate that， compared with models such as ASTGCN，Graph-WaveNet，and STTNs， the RMSE and MAE of the MRCSTGCN model are reduced by 19% and 22% respectively， verifying its effectiveness in spatio-temporal prediction of wave periods.

Keywords： wave period spatio-temporal prediction; graph convolution; multi-region collborative prediction; deep learning

波浪周期是表征波浪传播特性的核心参数，直接决定海面波动的频率与形态，其变化显著影响海面波动强度。（剩余9571字）