METHODS article
Front. Mar. Sci.
Sec. Physical Oceanography
A Significant Wave Height Prediction Method Combining VMD Decomposition and the GVSAO-CNN-BiGRU-SA Model
Zongquan Ying 1,2,3
Wengeng Shen 1,4,2,3
Xuegang Wang 5,2,3
Zhao Yiming 5,2,3
Meihong Lin 5,2,3
1. CCCC Fourth Harbor Engineering Institute Co., Ltd, Guangzhou, China
2. Southern Marine Science and Engineering Guangdong Laboratory - Zhuhai, Zhuhai, China
3. Key Laboratory of Environment and Safety Technology of Transportation Infrastructure Engineering, CCCC, Guangzhou, China
4. Dalian University of Technology State Key Laboratory of Coastal and Offshore Engineering, Dalian, China
5. CCCC Fourth Harbor Engineering Institute Co Ltd, Guangzhou, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
To improve the accuracy and robustness of significant wave height prediction under complex marine conditions, a multi-strategy Snow Ablation Optimization (GVSAO) model based on the Good Point Set Initialization Strategy (G), Cyclic Oscillation Mutation Strategy (V), and Snow Ablation Optimizer (SAO) is proposed to enhance parameter optimization. The GVSAO model combines Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BiGRU), and a Self-Attention Mechanism (SA) to construct the GVSAO-CNN-BiGRU-SA framework, which fully exploits the nonlinear characteristics of wave height time series. The study utilizes observed data from two observation points along the U.S. East Coast to the Gulf of Mexico (Stations 41013 and 42002) as well as from the Arabian Sea (Station 23020) and the Pacific Ocean (Station 46044). Comparative experiments on input feature combinations reveal that Intrinsic Mode Function (IMF) components derived from Variational Mode Decomposition (VMD) contribute more significantly to prediction accuracy than single physical features by effectively capturing dynamic time-frequency characteristics. The results demonstrate that the GVSAO model outperforms SAO, GSAO, and VSAO in terms of global exploration and stability, as validated by performance comparisons on the CEC2005 benchmark functions. Compared with the BiGRU model, the GVSAO-CNN-BiGRU-SA model exhibited superior performance, with RMSE reduced by 44.01% at Station 41013 and 15.12% at Station 42002. Similarly, it outperformed the CNN-BiGRU and CNN-BiGRU-SA models across all key metrics. The model achieved high-accuracy predictions in diverse marine environments, with relative mean errors within 0.5472%, RMSE within 0.1064 m, and correlation coefficients (R2) exceeding 0.99. Furthermore, in multi-step forecasting (3 to 48 hours), the model maintained high reliability with R2 values remaining above 0.84 across diverse geographic environments. The GVSAO-CNN-BiGRU-SA model provides a reliable solution for wave height prediction, contributing to marine engineering early warnings and energy utilization.
Summary
Keywords
CNN-BiGRU, CyclicOscillation Mutation Strategy, Good Point Set Initialization Strategy, Self-attention, Significant wave height prediction, snow ablation optimizer
Received
08 January 2026
Accepted
19 February 2026
Copyright
© 2026 Ying, Shen, Wang, Yiming and Lin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Wengeng Shen
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.