顾及时差特征的LSTM模型余水位短期预报  

Short-term prediction of residual water level by using LSTM model with regard to time difference

作　　者：冷建徽 许军[1] LENG Jianhui;XU Jun(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590

出　　处：《海洋测绘》2024年第2期27-30,共4页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(41501500)。

摘　　要：目前余水位预报都是采用单站方式,仅基于余水位的自相关性。针对较大范围的沿岸验潮站余水位预报,进一步结合余水位的空间强相关性和站间余水位的时差信息,以“预测站-辅助站”验潮站组的形式,由历史同步余水位数据训练多变量LSTM(long short-term memory)长短期记忆网络模型,实现预测站的余水位预报。渤海沿岸四个长期验潮站的实例分析表明:所提的预报方法因增加利用了时域上的时差信息,预报精度优于三类单站方法,并显著增大了预报时长。方法可用于解决大范围航海动态水位保障中的余水位预报问题。At present,the residual water level prediction adopts a single-station method,which is only based on the autocorrelation of the residual water level.Aiming at the prediction of residual water level at large-scale coastal tidal stations,this paper combines the spatial correlation and the time difference information of residual water levels between stations.In the form of a"reference station-auxiliary station"tidal gauge station group,we use historical synchronous residual water level data to train a multivariable LSTM(Long Short-Term Memory)neural network to achieve residual water level prediction at reference stations.The results of four long-term tidal stations along the Bohai coast shows that the prediction accuracy of the method proposed in this paper is better than that of the single station independent prediction method,which can significantly increase the prediction time length.The method proposed in this paper can forecast the residual water level in large-scale mautical dynamic water level assurance.

关 键 词：海洋潮汐 动态水位 余水位预报 长短期记忆网络模型 潮时差 

分 类 号：P229.7[天文地球—大地测量学与测量工程]