摘要
通过合成分析和回归分析,研究了影响辽东湾海冰变化的局地和大尺度环流因子,并基于一种深度学习方法——长短时记忆神经网络(LSTM),建立了辽东湾海冰延伸期预报模型。结果表明,LSTM模型能较好地预报出未来15 d辽东湾海冰的总体发展趋势、浮冰外缘线离岸距离的振荡变化及峰值发生时间等关键特征,1~15 d预报的平均绝对误差为4.1~5.7 n mile①,均方根误差为5.4~7.5 n mile。LSTM模型的预报时效可达到15 d,较目前海冰数值预报(5~7 d)的时效延长一倍,且运算速度极快,能够节省大量的计算资源和时间成本。该模型的建立为利用深度学习方法开展海洋和气象预报提供了一种新思路。
In this paper,the local and large-scale circulation factors affecting the variation of sea ice in the Liaodong Bay are studied by composite analysis and regress analysis,and then an extended-range forecast model of sea ice edge(SIE)in the Liaodong Bay is established based on a deep learning method-Long Short Term Memory Network(LSTM).The results show that the LSTM model can well predict the variation trend,generation and dissipation oscillation,peak and other key characteristics of the SIE in the Liaodong Bay in the next 15 days.The MAE varies from 4.1 to 5.7 n miles,and RMSE varies from 5.4 to 7.5 n miles,which is consistent to the sea ice numerical forecasting model.But the LSTM model extends the period of validity to 15 days,as twice as that of the numerical forecast(5~7).Besides,the operation speed is extremely fast,which can save a lot of computing resources and time cost.The establishment of this model provides a new idea for forecasting other marine and meteorological variables with deep learning method.
作者
焦艳
黄菲
高松
刘清容
冀承振
王宁
曹雅静
于清溪
JIAO Yan;HUANG Fei;GAO Song;LIU Qing-Rong;JI Cheng-Zhen;WANG Ning;CAO Ya-Jing;YU Qing-Xi(The Key Laboratory of Physical Oceanography,Ocean University of China,Qingdao 266100,China;North China Sea Marine Forecast Center,Qingdao 266061,China;North China Sea Data&Information Service,Qingdao 266061,China)
出处
《中国海洋大学学报（自然科学版）》
CAS
CSCD
北大核心
2020年第6期1-11,共11页
Periodical of Ocean University of China
基金
国家重点研究发展计划项目(2016YFC1402000,2015CB953904)
国家自然科学基金项目(U1706216
41575067)资助。
