| 基于变分模态分解和注意力机制的浪高预测 |
| |
| 引用本文： | 卢鹏,年圣全,邹国良,王振华,郑宗生.基于变分模态分解和注意力机制的浪高预测[J].海洋测绘,2021(2):34-39. |
| |
| 作者姓名： | 卢鹏 年圣全 邹国良 王振华 郑宗生 |
| |
| 作者单位： | 上海海洋大学 信息学院,上海 201316 |
| |
| 基金项目： | 上海市地方能力建设项目(19050502100)；上海海洋大学科技发展专项(A2-2006-20-200211) |
| |
| 摘 要： | 针对海洋中的海浪高度数据存在非线性和非平稳性的特点,海浪高度的预测就变得相对复杂。基于变分模态分解(VMD),在引入注意力机制(AM)的基础上,对传统长短期记忆(LSTM)神经网络算法进行了改进,提出了一种基于混合模型的海浪高度预测算法。算法通过预处理、预测和重构3个主要步骤,对海浪高度的时间序列进行预测。为了比较和说明,以太平洋东北海盆海域和马尾藻海域的4个站点浮标数据进行实验。实验结果表明,本文提出的混合模型(VALM)将海浪高度数据分解为更平稳和更规则的子序列;可以更好的区分数据之间的重要程度,并能够携带更多信息的数据;与支持向量回归(SVR)、人工神经网络(ANN)和LSTM等模型进行比较,VALM模型的预测效果最好且具备一定的鲁棒性。
|
| 关 键 词： | 海浪高度预测 变分模态分解 注意力机制 长短期记忆神经网络 混合模型 |
Wave height prediction based on variational mode decomposition and attention mechanism |
| |
| LU Peng,NIAN Shengquan,ZOU Guoliang,WANG Zhenhu,ZHENG Zongsheng.Wave height prediction based on variational mode decomposition and attention mechanism[J].Hydrographic Surveying and Charting,2021(2):34-39. |
| |
| Authors: | LU Peng NIAN Shengquan ZOU Guoliang WANG Zhenhu ZHENG Zongsheng |
| |
| Institution: | College of Information Technology,Shanghai Ocean University,Shanghai 201316 ,China |
| |
| Abstract: | In view of the non-linearity and nonstationarity of wave height data in the ocean,the prediction of wave height becomes relatively complex.In this paper,based on Variational Mode Decomposition (VMD) and the introduction of Attention Mechanism (AM),the traditional LSTM neural network algorithm is improved,and a hybrid model based wave height prediction algorithm is proposed.The algorithm predicts the time series of wave height through three main steps: preprocessing,prediction and reconstruction.In order to compare and explain,the data of four station buoys in the northeast Pacific basin and the Sargasso Sea are used for experiments.The experimental results show that the proposed hybrid model (VALM) decomposes the wave height data into more stable and regular subsequences,and that can better distinguish the importance of data and carry more information;Compared with support vector regression (SVR),artificial neural network (ANN) and LSTM,the prediction effect of VALM model is the best and has certain robustness. |
| |
| Keywords: | |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《海洋测绘》浏览原始摘要信息 |
| 点击此处可从《海洋测绘》下载免费的PDF全文 |
|
