海洋测绘Issue(4)：37-42,6.DOI:10.3969/j.issn.1671-3044.2025.04.008
基于改进图卷积的多站点海浪高度预测方法
A multi-site wave height prediction method based on improved graph convolutional network
摘要
Abstract
The variation of ocean wave height exhibits not only temporal dynamics but also spatial influences from surrounding maritime areas.Existing methods are predominantly limited to extracting temporal features from individual stations,neglecting the critical need to capture inter-site spatio-temporal dependencies of wave heights within the same region,this thesis proposes a SD-STSGCN for multi-site wave height prediction.First,density-based K-means clustering is employed to group monitoring stations.Second,a scaled distance factor is introduced to construct an adaptive adjacency matrix that dynamically adjusts inter-station connection weights.Finally,a spatio-temporally synchronous graph convolution module integrated with dilated convolution operations captures heterogeneous spatio-temporal dependencies,followed by nonlinear mapping to output predicted wave heights for future time periods across grouped stations.Extensive regional experiments conducted across 44 multi-dimensional maritime stations demonstrate that SD-STSGCN outperforms benchmark models(e.g.,LSTM and TCN)in prediction accuracy.The proposed method effectively exploits multi-site spatio-temporal correlations,providing a valuable supplementary approach for ocean wave height forecasting.关键词
海浪高度预测/多站点预测/时空同步图卷积/时空相关性/邻接矩阵Key words
wave height prediction/multi-site prediction/spatial-temporal synchronous graph convolutional network/spatial-temporal correlation/adjacency matrix分类
天文与地球科学引用本文复制引用
卢鹏,王慧,王振华,郑宗生..基于改进图卷积的多站点海浪高度预测方法[J].海洋测绘,2025,(4):37-42,6.基金项目
上海市科委科研计划项目(20dz1203800). （20dz1203800）
