热带海洋学报 ›› 2024, Vol. 43 ›› Issue (4): 76-85.doi: 10.11978/2023113CSTR: 32234.14.2023113

• 海洋水文学 • 上一篇    下一篇

融合MIC与Res-LSTM模型的有效波高预测

朱道恒1(), 李彦2, 李志强1(), 刘润3   

  1. 1.电子与信息工程学院, 广东海洋大学, 广东 湛江 524088
    2.大数据与信息工程学院, 贵州大学, 贵州 贵阳 550025
    3.化学与环境学院, 广东海洋大学, 广东 湛江 524088
  • 收稿日期:2023-08-04 修回日期:2023-10-11 出版日期:2024-07-10 发布日期:2024-07-22
  • 作者简介:

    朱道恒(1992—), 男, 湖北省孝感市人, 博士, 从事海岸带与海洋监测技术研究。email:

  • 基金资助:
    国家自然科学基金项目(42176167); 广东海洋大学科研启动经费项目(060302112317)

Fusing MIC and Res-LSTM models for significant wave height prediction

ZHU Daoheng1(), LI Yan2, LI Zhiqiang1(), LIU Run3   

  1. 1. School of Electronics and Information Engineering, Guangdong Ocean University, Zhanjiang 524088, China
    2. School of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China
    3. School of Chemistry and Environment, Guangdong Ocean University, Zhanjiang 524088, China
  • Received:2023-08-04 Revised:2023-10-11 Online:2024-07-10 Published:2024-07-22
  • Supported by:
    National Natural Science Foundation of China(42176167); Scientific Research Startup Foundation of Guangdong Ocean University(060302112317)

摘要:

有效波高 (significant wave height, SWH)的预测在海洋运输和海上活动方面发挥着重要作用。基于中国阳江海陵岛近岸实测数据, 提出一种融合最大信息系数 (maximal information coefficient, MIC)、残差网络 (residual network, ResNet)和长短期记忆网络 (long short-term memory networks, LSTM)的预测模型。首先, 采用MIC算法从数据集中筛选出与预测指标相关性高的参数作为模型的输入; 然后将ResNet引入LSTM中, 构建Res-LSTM预测模型; 最后选择相关系数 (r-squared, R2)、均方根差(root mean squared error, RMSE)、平均绝对误差(mean absolute error, MAE)和平均绝对百分比误差(mean absolute percentage error, MAPE)来评价预测结果。同时, 对比了XGBoost (extreme gradient boosting)、SVR(support vector regression)和LSTM网络的预测效果。结果表明, MIC-Res-LSTM模型能够提高短时有效波高预测值的精度。

关键词: 波高预测, 最大信息系数, 残差网络, 长短期记忆网络, 支持向量回归

Abstract:

The prediction of significant wave height (SWH) plays an important role in marine transportation and maritime activities. Based on the near-shore real measurement data of the Hailing Island, Yangjiang, China, a network model integrating the maximum information coefficient algorithm (MIC), residual network (ResNet) and long and short-term memory network (LSTM) is proposed. Firstly, the MIC algorithm was used to screen out the parameters with high correlation with the target predictors from the dataset as the input of the model. Then the residual network was introduced into the LSTM to construct the Res-LSTM prediction model. Finally, the r-squared (R2), root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) were selected to evaluate the prediction results. Meanwhile, the prediction results of extreme gradient boosting (XGBoost) network, support vector regression (SVR) network and LSTM network were compared. The results demonstrate that the MIC-Res-LSTM model can improve the accuracy of the short-time significant wave height prediction values.

Key words: wave height prediction, maximum information coefficient, residual network, long and short-term memory network, support vector regression