基于生物认知启发的有效波高预测模型

韩坤, 陈鑫

清华大学学报(自然科学版) ›› 2026, Vol. 66 ›› Issue (11) : 2428-2440.

PDF(4721 KB)
PDF(4721 KB)
清华大学学报(自然科学版) ›› 2026, Vol. 66 ›› Issue (11) : 2428-2440. DOI: 10.16511/j.cnki.qhdxxb.2026.27.062
水利水电工程

基于生物认知启发的有效波高预测模型

  • 韩坤, 陈鑫
作者信息 +

Bio-cognition-inspired significant wave height prediction model

  • HAN Kun, CHEN Xin
Author information +
文章历史 +

摘要

准确预测有效波高,对波浪发电和海上活动至关重要。为克服传统模型在处理波浪随机性和非线性方面的局限性,该文提出一种基于生物认知启发的预测模型。该模型包括3个模块:1)感官捕获模块,采用完全集合经验模态分解自适应噪声(CEEMDAN)将复杂序列分解为易于建模的组成;2)大脑分析模块,结合双向长短期记忆网络(BiLSTM)和改进的逆Transformer(TSSA-iTransformer)模拟大脑左右半球,捕捉时间依赖关系并建模多变量间的全局相关性;3)预测模块,通过连接融合生成有效波高预测。预测结果表明,在不同海域的数据集上,所提模型在1 h、6 h、12 h和24 h的有效波高预测中均表现出优异的精度和稳定性,且在极端海况下仍保持稳健性,优于基线模型和现有代表性模型。该研究为复杂海洋环境中的有效波高预测提供了一种有效的解决方案。

Abstract

[Objective]Accurate and reliable prediction of significant wave height (SWH) is of great importance for achieving sustainable wave energy development and ensuring the safety of offshore engineering operations. Due to the pronounced nonlinearity, non-stationarity, and randomness of SWH, traditional wave height prediction models often fail to fully capture the complex nonlinear characteristics embedded in SWH sequences, resulting in limited predictive accuracy. To overcome these limitations, this study draws inspiration from the superior performance of biomimetic deep learning models in time-series forecasting tasks and proposes a biologically cognition-inspired SWH prediction model. [Methods] The proposed model follows the biological cognitive information-processing paradigm and consists of three functionally complementary and collaboratively operating modules. First, the sensory capture module simulates the preprocessing mechanism of biological sensory systems by employing Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to adaptively decompose the original non-stationary SWH sequence into multiple intrinsic mode functions at different temporal scales, thereby effectively alleviating mode mixing and large residual noise problems. Second, the brain analysis module mimics the functional specialization of the human left and right hemispheres. The right hemisphere focuses on sequential perception and utilizes Bidirectional Long Short-Term Memory (BiLSTM) to extract long-term and short-term temporal dependencies in wave evolution. The left hemisphere employs an improved inverted Transformer (TSSA-iTransformer) to model global correlations among multivariate features. In this framework, Token Statistics Self-Attention (TSSA) replaces the standard self-attention mechanism, significantly reducing computational burden and memory consumption while preserving global dependency modeling capability. Furthermore, the decomposed components generated by CEEMDAN are independently fed into the parallel left and right hemisphere branches, thereby avoiding feature interference commonly observed in traditional serial fusion architectures. Finally, the prediction module simulates the information fusion mechanism of the higher cerebral cortex by integrating the global temporal features and multivariate correlation features extracted from the brain analysis module and generating the final SWH prediction through fully connected layers. [Results] Experimental results on real-world buoy datasets demonstrate that: 1) The proposed model outperforms all baseline and existing models in the 1 h forecasting task, with ΔP remaining below 10.37%, demonstrating strong capability in accurately capturing instantaneous wave variations; 2) In the 6 h, 12 h, and 24 h multi-step forecasting tasks, the model maintains stable performance and effectively mitigates the error accumulation problem commonly observed in traditional long-term prediction models; 3) In terms of computational efficiency, the single-sample inference time of the proposed model is improved by an average of 44.5% compared with similar biomimetic models, demonstrating its engineering potential for real-time ocean forecasting applications; 4) Ablation experiments further validate the rationality and necessity of each module design. In addition, the model demonstrates strong prediction stability under different sea conditions and achieves excellent cross-scenario generalization performance on two additional independent buoy stations. [Conclusions] Overall, the biologically cognition-inspired architectural design provides a systematic and interpretable framework for SWH prediction. By effectively integrating sequence decomposition, temporal dependency modeling, and global multivariate correlation learning, the proposed model achieves accurate, stable, and efficient SWH prediction across different marine environments, thereby providing important technical support for marine disaster warning and ocean environmental monitoring.

关键词

波高预测 / 仿生模型 / 深度学习 / 双向长短期记忆网络 / Transformer

Key words

wave height prediction / biomimetic model / deep learning / bidirectional long short-term memory / Transformer

引用本文

导出引用
韩坤, 陈鑫. 基于生物认知启发的有效波高预测模型[J]. 清华大学学报(自然科学版). 2026, 66(11): 2428-2440 https://doi.org/10.16511/j.cnki.qhdxxb.2026.27.062
HAN Kun, CHEN Xin. Bio-cognition-inspired significant wave height prediction model[J]. Journal of Tsinghua University(Science and Technology). 2026, 66(11): 2428-2440 https://doi.org/10.16511/j.cnki.qhdxxb.2026.27.062
中图分类号: P743.2   

参考文献

[1] SHADMANI A, NIKOO M R, GANDOMI A H, et al. A review of machine learning and deep learning applications in wave energy forecasting and WEC optimization [J]. Energy Strategy Reviews, 2023, 49: 101180.
[2] ALI M, PRASAD R, XIANG Y, et al. Near real-time significant wave height forecasting with hybridized multiple linear regression algorithms [J]. Renewable and Sustainable Energy Reviews, 2020, 132: 110003.
[3] MARKUSZEWSKI P, KOSECKI S, PETELSKI T. Sea spray aerosol fluxes in the Baltic Sea region: Comparison of the WAM model with measurements [J]. Estuarine, Coastal and Shelf Science, 2017, 195: 16-22.
[4] ZHENG Z H, ALI M, JAMEI M, et al. Multivariate data decomposition based deep learning approach to forecast one-day ahead significant wave height for ocean energy generation [J]. Renewable and Sustainable Energy Reviews, 2023, 185: 113645.
[5] 魏双, 安毅, 余向军, 等. 基于Non stationary-CNN- Transformer的海浪有效波高预测[J]. 太阳能学报, 2024, 45(10): 673-682. WEI S, AN Y, YU X J, et al. Effective wave height prediction based on non stationary-CNN-transformer for ocean waves [J]. Acta Energiae Solaris Sinica, 2024, 45(10): 673-682. (in Chinese)
[6] POKHREL P, IOUP E, SIMEONOV J, et al. A transformer-based regression scheme for forecasting significant wave heights in oceans[J]. IEEE Journal of Oceanic Engineering, 2022, 47(4): 1010-1023.
[7] SADEGHIFAR T, NOURI M M, TORABI A M, et al. Coastal wave height prediction using Recurrent Neural Networks (RNNs) in the south Caspian Sea [J]. Marine Geodesy, 2017, 40(6): 454-465.
[8] 刘东林, 姚朝帮, 余嘉威, 等. 基于LSTM-Transformer的波高反演研究[J]. 华中科技大学学报(自然科学版), 2026, 54(4): 1-6. LIU D L, YAO C B, YU J W, et al. Research on wave height inversion based on LSTM-Transformer [J]. Journal of Huazhong University of Science & Technology (Natural Science Edition), 2026, 54(4): 1-6. (in Chinese)
[9] 庞军恒, 黄炜楠, 董胜. 基于多变量DSD-LSTM模型的有效波高预测[J]. 太阳能学报, 2024, 45(7): 121-127. PANG J H, HUANG W N, DONG S. Prediction of significant wave height based on multivariable DSD-LSTM model [J]. Acta Energiae Solaris Sinica, 2024, 45(7): 121-127. (in Chinese)
[10] 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(Pt 2): 112747.
[11] CHEN J X, LI S B, ZHU J Z, et al. Significant wave height prediction based on variational mode decomposition and dual network model [J]. Ocean Engineering, 2025, 323: 120533.
[12] SAREEN K, PANIGRAHI B K, SHIKHOLA T, et al. An integrated decomposition algorithm based bidirectional LSTM neural network approach for predicting ocean wave height and ocean wave energy [J]. Ocean Engineering, 2023, 281: 114852.
[13] LAKE B M, BARONI M. Human-like systematic generalization through a meta-learning neural network [J]. Nature, 2023, 623(7985): 115-121.
[14] TEMENG V A, ZIGGAH Y Y, ARTHUR C K. A novel artificial intelligent model for predicting air overpressure using brain inspired emotional neural network [J]. International Journal of Mining Science and Technology, 2020, 30(5): 683-689.
[15] WU H, LIANG Y, GAO X Z. Left-right brain interaction inspired bionic deep network for forecasting significant wave height [J]. Energy, 2023, 278: 127995.
[16] WU H, GAO X Z, HENG J N. Bio-multisensory-inspired gate-attention coordination model for forecasting short-term significant wave height [J]. Energy, 2024, 294: 130887.
[17] WANG J F, CHEN Y, DONG Z K, et al. SABV-depth: A biologically inspired deep learning network for monocular depth estimation [J]. Knowledge-Based Systems, 2023, 263: 110301.
[18] WANG T C, ZHANG J W, XIONG J J, et al. VisuaLNet: An end-to-end human visual system inspired framework to reduce inference latency of deep neural networks [J]. IEEE Transactions on Computers, 2022, 71(11): 2717-2727.
[19] TORRES M E, COLOMINAS M A, SCHLOTTHAUER G, et al. A complete ensemble empirical mode decomposition with adaptive noise [C]//2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Prague, Czech Republic: IEEE, 2011: 4144-4147.
[20] LAKATOS P, CHEN C M, O'CONNELL M N, et al. Neuronal oscillations and multisensory interaction in primary auditory cortex [J]. Neuron, 2007, 53(2): 279-292.
[21] CHEN G, KANG B, LINDSEY J, et al. Modularity and robustness of frontal cortical networks [J]. Cell, 2021, 184(14): 3717-3730.e24.
[22] YIN X X, WANG Y, LI J J, et al. Lateralization of short-term memory in the frontal cortex [J]. Cell Reports, 2022, 40(7): 111190.
[23] LIU Y, HU T G, ZHANG H R, et al. iTransformer: Inverted transformers are effective for time series forecasting [J]. arXiv preprint arXiv: 2310.06625, 2023.
[24] WU Z Y, DING T J, LU Y F, et al. Token statistics transformer: Linear-time attention via variational rate reduction [J]. arXiv preprint arXiv: 2412.17810, 2024.
[25] TANG H D, KANG F, LI X Y, et al. Short-term photovoltaic power prediction model based on feature construction and improved transformer [J]. Energy, 2025, 320: 135213.
[26] SHEN X N, SHEN J J, ZHANG Y T, et al. A multi-source feature engineering-enhanced framework for mid-to-long-term EV charging load forecasting: Integrating self-adaptive optimization and BiLSTM-iTransformer predictor [J]. Energy, 2025, 339: 139020.
[27] 刘新宇, 潘宇, 王亚辉, 等. 增强的超短期风电功率预测: 一种PatchTST-POA-VMD-iTransformer混合模型[J]. 电力系统保护与控制, 2025, 53(19): 68-78. LIU X Y, PAN Y, WANG Y H, et al. Enhanced ultra-short-term wind power forecasting: a PatchTST-POA-VMD-iTransformer hybrid model [J]. Power System Protection and Control, 2025, 53(19): 68-78.
[28] WU Y N, WANG J T, ZHANG R F, et al. RIME-CNN-BiLSTM: A novel optimized hybrid enhanced model for significant wave height prediction in the Gulf of Mexico [J]. Ocean Engineering, 2024, 312(Pt 2): 119224.
[29] HUANG Q, CUI Z D. Study on prediction of ocean effective wave height based on hybrid artificial intelligence model [J]. Ocean Engineering, 2023, 289(Pt 1): 116137.
[30] WANG L N, WANG X D, DONG C M, et al. Wave predictor models for medium and long term based on dual attention- enhanced Transformer [J]. Ocean Engineering, 2024, 310(Pt 2): 118761.
[31] SU Z H, CHEN H L, YANG C. Hybrid framework of deep extreme learning machine (DELM) based on sparrow search algorithm for non-stationary wave prediction [J]. Ocean Engineering, 2024, 311(Pt 2): 118993.
[32] FANG L, WU W M, CAO F F, et al. Biological visual-cognition- inspired deep network for short-term significant wave height prediction [J]. Engineering Applications of Artificial Intelligence, 2025, 162: 112607.
[33] ZHU T B, CAI F, CHEN D D. Robust offshore wind power forecasting under extreme marine conditions using multi-source feature fusion and Kolmogorov-Arnold networks [J]. Journal of Marine Science and Engineering, 2026, 14(6): 573.

基金

国家自然科学基金面上项目(12472415)

PDF(4721 KB)

Accesses

Citation

Detail

段落导航
相关文章

/

〈 〉