复杂的海洋环境下，船舶航行安全受波高显著影响，波高与船舶运动状态密切相关，需构建基于深度学习的波高预测模型以保障航行安全。由于海浪波高具有非平稳的特性，传统预测方法效果欠佳。本文创新提出基于灰狼优化算法下最优参数的海浪波高特征预报复合模型，采取灰狼算法（Grey Wolf Optimization，GWO）优化变分模态分解（Variational Mode Decomposition，VMD）参数中模态分量数K和惩罚因子$ \alpha $处理数据后提取波高序列固有模态函数（Intrinsic Mode Functions，IMF），以排列熵（Permutation Entropy，PE）为标准筛选信号，并将有效的模态分量作为长短时记忆神经网络（Long-Short Term Memory Network，LSTM）模型输入；构建VMD-PE-GWO-LSTM复合模型。经南海实测数据验证，该模型将波高数据分解为8个趋势项序列，预报精度达到R2=0.9857，能更精准地预报波高数据从而保障航行中的船舶安全。
Under complex marine conditions, wave height significantly impacts ship navigation safety and is closely correlated with vessel motion states. Establishing a deep learning-based wave height prediction model is therefore essential for ensuring navigational security. Given the non-stationary nature of ocean waves, traditional forecasting methods often yield suboptimal results. This study innovatively proposes a composite wave height prediction model based on optimal parameters derived from the Grey Wolf Optimization (GWO) algorithm. The methodology employs GWO to optimize key parameters in Variational Mode Decomposition (VMD)—specifically, the number of mode components (K) and the penalty factor ($ \alpha $). After processing the data, Intrinsic Mode Functions (IMFs) are extracted from wave height sequences, with Permutation Entropy (PE) serving as the criterion for signal screening. Effective modal components are subsequently fed into a Long Short-Term Memory (LSTM) neural network, culminating in the construction of the VMD-PE-GWO-LSTM composite model. Validated against in-situ measurements from the South China Sea, this model decomposes wave height data into eight trend components and achieves a prediction accuracy of R2=0.9857, demonstrating enhanced precision in wave height forecasting to safeguard vessel safety during navigation.
2026,48(3): 46-52 收稿日期：2025-5-16
DOI：10.3404/j.issn.1672-7649.2026.03.007
分类号：U66；P731.33
基金项目：国家自然科学基金资助项目（11702059）
作者简介：冷春花(1999-),女,硕士,研究方向为时域与频域特征融合的波浪预测
参考文献：
[1] LI G G, ZHANG H, LV T, et al. Regional significant wave height forecast in the East China Sea based on the Self-Attention ConvLSTM with SWAN model [J]. Ocean Engineering, 2024, 312.
[2] YAO J, WU W H. Wave height forecast method with multi-step training set extension LSTM neural network [J]. OCEAN ENGINEERING, 2022, 263.
[3] 陈恒轩, 张雷, 杜传顺, 等. 基于MAF-GWO-LSTM算法的海浪有义波高预测模型[J]. 舰船科学技术, 2024, 46(21): 33-39
CHEN H X, ZHANG L, DU C S, et al. Prediction model of significant wave height based on MAF-GWO-LSTM algorithm[J]. Ship Science and Technology, 2024, 46(21): 33-39
[4] LUO Q R, XU H, BAI L. H. Prediction of significant wave height in hurricane area of the Atlantic Ocean using the Bi-LSTM with attention model [J]. Ocean Engineering, 2022, 266.
[5] SUN X W, LU T D, HU S Q, et al. A new algorithm for predicting dam deformation using grey wolf-optimized variational mode long short-term neural network [J]. Remote Sensing, 2024, 16(21): 3978.
[6] DING T, WU D A, SHEN L S, et al. Prediction of significant wave height using a VMD-LSTM-rolling model in the South Sea of China [J]. Frontiers in Marine Science, 2024, 11.
[7] SASUBILLI S, VERMA M. InFi-BERT 1.0: transformer-based language model for indian financial volatility prediction [M]. Machine learning and Principles and Practice of Knowledge Discovery in Databases, Ecml Pkdd 2022, PT II. 2023: 128-138.
[8] LIU D W, CHEN H F, TANG Y, et al. Slope micrometeorological analysis and prediction based on an ARIMA model and data-fitting system [J]. Sensors, 2022, 22(3).
[9] SHANG S, HE K N, WANG Z B, et al. Sea clutter suppression method of HFSWR based on RBF neural network model optimized by improved GWO algorithm [J]. Computational Intelligence and Neuroscience, 2020.
[10] SHEN W E, YING Z Q, ZHAO Y M, et al. Significant wave height prediction in monsoon regions based on the VMD-CNN-BiLSTM model [J]. Frontiers in Marine Science, 2024, 11.
[11] ANAND P, JAIN S, SAVALIYA H. New improved wave hybrid models for short-term significant wave height foreCasting[J]. IEEE ACCESS, 2023, 11: 109841-109855.
