| 基于LSTM的海洋表面短期风速预测研究 |
| |
| 引用本文： | 李亚蒙,孙宝楠,丁军航,官晟.基于LSTM的海洋表面短期风速预测研究[J].海洋科学,2022,46(11):55-66. |
| |
| 作者姓名： | 李亚蒙 孙宝楠 丁军航 官晟 |
| |
| 作者单位： | 青岛大学自动化学院, 山东 青岛 266071;自然资源部第一海洋研究所, 山东 青岛 266061;自然资源部第一海洋研究所, 山东 青岛 266061;自然资源部海洋环境科学与数值模拟重点实验室, 山东 青岛 266061;山东省海洋环境科学与数值模拟重点实验室, 山东 青岛 266061;青岛海洋科学与技术试点国家实验室区域海洋动力学与数值模拟功能实验室, 山东 青岛 266237;青岛大学自动化学院, 山东 青岛 266071;山东省工业控制技术重点实验室, 山东 青岛 266071;青岛大学山东省生态纺织协同创新中心, 山东 青岛 266071 |
| |
| 基金项目： | 科技部重大科学仪器专项项目(2018YFF01014100) |
| |
| 摘 要： | 为实现对海面风速精确的短期预测,提出了一种基于长短期记忆(LSTM,long short-term memory)神经网络的短期风速预测模型,选取OceanSITES数据库中单个浮标站点采集的风速历史数据作为模型输入,经过训练设置最佳参数等步骤,实现了以LSTM方法,对该站点所在海区海面风速在各季节性代表月份海面风速的24h短期预测。同时通过不同预测时长的实验以及与BP(back propagation)神经网络神经网络和径向基函数神经网络(radial basis function neural network,RBF)的预测效果对比实验,证明了LSTM预测方法相比上述两种神经网络预测方法,在海表面风速预测应用中的优越性。最后通过多个海域对应的站点风速数据预测实验,证明了LSTM神经网络模型的普遍适用性,由相关系数和预测误差的分析可知该方法具备应对急剧变化数据的预测稳定性,可以作为海洋表面风速短期预测的一种可靠方法。
|
| 关 键 词： | 神经网络 长短期记忆网络模型 海面风速 短期预测 |
| 收稿时间： | 2021/11/16 0:00:00 |
| 修稿时间： | 2022/1/23 0:00:00 |
Short-term wind-speed prediction of ocean surface based on LSTM |
| |
| LI Ya-meng,SUN Bao-nan,DING Jun-hang,GUAN Sheng.Short-term wind-speed prediction of ocean surface based on LSTM[J].Marine Sciences,2022,46(11):55-66. |
| |
| Authors: | LI Ya-meng SUN Bao-nan DING Jun-hang GUAN Sheng |
| |
| Institution: | School of Automation, Qingdao University, Qingdao 266071, China;First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China;First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China;Key Laboratory of Marine Science and Numerical Modeling, Ministry of Natural Resources, Qingdao 266061, China;Shandong Key Laboratory of Marine Science and Numerical Modeling, Qingdao 266061, China;Laboratory for Regional Oceanography and Numerical Modeling, Pilot National Laboratory for Marine Science and Technology(Qingdao), Qingdao 266237, China;School of Automation, Qingdao University, Qingdao 266071, China;Shandong Key Laboratory of Industrial Control Technology, Qingdao 266071, China;Collaborative Innovation Center for Eco-Textiles of Shandong Province, Qingdao University, Qingdao 266071, China |
| |
| Abstract: | Sea surface wind speed plays a critical role in production and operation activities such as marine space launches and wind power deployment. However, because sea surface wind speed is highly nonlinear and stochastic, it is relatively challenging to estimate precisely. A short-term wind-speed prediction model based on long short-term memory (LSTM) neural network is suggested to accomplish reliable short-term prediction of sea surface wind speed. The historical wind speed data gathered by a single buoy station in the OceanSITES database is chosen as the model input, and the LSTM techniques are implemented by training the best parameters. Using this LSTM method, a 24-hour short-term forecast of the sea surface wind speed where the station is located in each seasonal representative month is realized. Simultaneously, through experiments involving various prediction durations and comparison experiments of prediction effects with back propagation neural network and radial basis function neural network, it is demonstrated that the LSTM prediction approach is superior to the above two neural network prediction methods in the application of sea surface wind-speed prediction. Finally, the LSTM neural network model is demonstrated to be globally applicable to wind speed prediction experiment data at stations representing diverse marine areas. The analysis of correlation coefficient and prediction error reveals that this method has the prediction stability to deal with quickly changing data and can be employed as a reliable method for short-term prediction of ocean surface wind speed. |
| |
| Keywords: | neural network LSTM network model sea surface wind speed short-term prediction |
|
| 点击此处可从《海洋科学》浏览原始摘要信息 |
| 点击此处可从《海洋科学》下载免费的PDF全文 |
|
