| 摘要: |
| 海浪直接影响海上活动和航行安全,同时也蕴藏着巨大的可再生能源,对海浪核心参数之一波高预测至关重要。基于2015年7月~2022年6月山东小麦岛(36°N, 120.6°E)站点实测的波高数据, 利用反向传播神经网络(back-propagation neural network, BPNN)、长短记忆网络(long short-term memory, LSTM)和支持向量机回归(support vector regression, SVR)三种机器学习模型对波高进行预测, 并分析了瑞利参数的引入对预测结果的影响。结果显示, 模型输入项引入瑞利参数后, 对1 h和6 h波高预测提升效果有限, 预测值与测试集的相关性提升不超过0.02, 均方根误差的降低不超过0.01 m; 在12 h和24 h的预测中, BPNN和LSTM模型预测结果相关性提升0.03~0.07, 均方根误差降低0.02~0.03 m, 而SVR模型预测结果变化不显著。说明瑞利参数有助改善BPNN和LSTM模型中长期海浪预报。此外, 特征扰动方法(机器学习中特征重要性的计算方法之一)验证了瑞利参数在波高预测中的重要性, 瑞利参数的引入为波高的机器学习预测提供了新思路。 |
| 关键词: 波高 反向传播神经网络 长短记忆网络 支持向量机 机器学习 瑞利参数 |
| DOI：10.11693/hyhz20230900180 |
| 分类号:P731 |
| 基金项目:国家重点研发课题,2022YFC3104805号;国家自然科学基金项目,42276019号。 |
附件 |
|
| THE APPLICATION OF THE RAYLEIGH PARAMETER IN MACHINE LEARNING PREDICTION OF WAVE HEIGHT |
|
HU Ming-Hao1, XIE Ling-Ling1,2,3, LI Ming-Ming1,2,3, LIANG Peng1,2,3
|
|
1.Laboratory of Coastal Ocean Variability and Disaster Warning, College of Oceanography and Meteorology, Guangdong Ocean University, Zhanjiang 524088, China;2.Key Laboratory of Shelf and Deep-sea Environment and Resources of Guangdong Higher Education Institutes, Zhanjiang 524088, China;3.Key Laboratory of Spatial Ocean Remote Sensing and Application, Ministry of Natural Resources, Zhanjiang 524088, China
|
| Abstract: |
| Waves directly affect maritime activities and navigation safety, and also contain enormous renewable energy, making it crucial to predict wave height, one of the core parameters of waves. This study is based on wave height data measured at the Xiaomai Island Station (36°N, 120.6°E) in Shandong Province from July 2015 to June 2022. Three machine learning models, back-propagation neural network (BPNN), long short-term memory network (LSTM), and support vector machine regression (SVR), were used to predict wave heights, and the influence of Rayleigh parameter introduction into the prediction results was analyzed. The results show that the introduction of the Rayleigh parameter as one of the input features had limited improvement on the prediction of wave heights of 1 h and 6 h forecasts. The correlation between the predicted values and the test dataset was no more than 0.02, and the reduction in root mean square error (RMSE) did not exceed 0.01 m. In the 12 and 24 h predictions, the correlation between the BPNN and LSTM models was improved by 0.03~0.07, and the RMSE was decreased by 0.02~0.03 m, while the SVR model did not show significant changes in the prediction results. Therefore, the Rayleigh parameter involvement can help improve the mid- to long-term wave forecasting in the BPNN and LSTM models. In addition, the feature perturbation method (one of the calculation methods of feature importance in machine learning) has verified the importance of the Rayleigh parameter in wave height prediction. The introduction of the Rayleigh parameter provided a new approach for machine learning prediction of wave heights. |
| Key words: wave height back-propagation neural network long short-term memory network support vector machine regression machine learning Rayleigh parameter |
