| 摘要: |
| 随着计算机技术和观测手段的提升,海浪预报技术与方法在不断进步,基于人工智能技术的海浪智能预报得到了很好的发展。近年来,深度学习作为人工智能主要的实现方式在海浪预测中得到了广泛运用。本文基于海浪智能预测的基本深度学习模型,总结了它们在海浪有效波高及海浪谱预测的应用。特别地,在海浪有效波高预测中,分别从基于时间序列预处理和LSTM的海浪单点预测模型、Conv-LSTM、注意力机制及序列到序列的连续预测模型4个方面对已有的海浪预测研究进行概述。通过以上总结,明确了深度学习模型在海浪预测发展中的优势以及存在的不足,并针对存在的问题提出了解决方案。最后,对未来海浪智能预测研究进行了展望。尽管在海浪理论的发展过程中深度学习方法不能完全替代海浪数值模式,但深度学习模型在未来的发展中能够更好地学习海浪时空特征,为建立海洋智能大模型提供指导价值,实现“海洋数字孪生”。 |
| 关键词: 深度学习 神经网络 海浪预测 海浪谱 |
| DOI：10.11759/hykx20240318001 |
| 分类号:P731.33 |
| 基金项目:汕头大学科研启动经费项目(NTF21036),海洋灾害预警与防护广东省重点实验室开放基金课题资助(GPKLMD2023005) |
|
| Research development about applications of deep learning in ocean wave |
|
JIANG Longyu1, HUA Feng1, JIANG Xingjie2, JIN Quan1, WANG Zeyu1
|
|
1.Guangdong Provincial Key Laboratory of Marine Disaster Prediction and Prevention, Shantou University, Shantou 515063, China;2.First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China
|
| Abstract: |
| With advancements in computer technology and observational methods, ocean wave prediction has progressed significantly, particularly through the use of artificial intelligence. Deep learning, a key component of artificial intelligence, has been widely applied to ocean wave prediction. This paper reviews the applications of deep learning models in predicting significant wave heights and wave spectra. It specifically focuses on four areas: single-point models using time-series preprocessing and long short-term memory (LSTM), ConvLSTM, attention mechanisms, and sequence-to-sequence models for continuous wave height predictions. The review outlines the advantages and disadvantages of deep learning models in ocean wave prediction and proposes solutions to current challenges. Finally, future research regarding wave prediction is summarized. Although deep learning cannot entirely replace numerical ocean wave models in theoretical development, they are poised to enhance our understanding of spatial and temporal characteristics of ocean waves. This should guide the creation of intelligent big ocean models and the realization of digital twin oceans. |
| Key words: deep learning neural network ocean wave prediction wave spectrum |
