海洋测绘2024，Vol.44Issue(4)：48-53,6.DOI:10.3969/j.issn.1671-3044.2024.04.011
基于ConvLSTM网络的北极海冰时空序列预测研究
Research on Arctic sea ice spatiotemporal sequence prediction based on convolutional long short term memory network
摘要
Abstract
Accurate prediction of sea ice concentration(SIC)is particularly crucial for opening Arctic shipping routes,supporting polar scientific research and resource development.In the field of ocean forecasting,statistical forecasting plays a vital role.This study introduces a cascaded Convolutional Long Short-Term Memory neural network(ConvLSTM)for medium-and short-term prediction of Arctic Sea ice concentration.This network has the ability of image processing and spatiotemporal prediction,which can accurately predict the spatiotemporal sequences of sea ice.It can handle input sequences of different lengths and demonstrates strong predictive potential in various data contexts.By optimizing the network architecture,it achieves stronger performance,accurately capturing and analyzing dynamic changes in SIC.Experimental results show that this model achieves a root mean square error of 0.0599 and a correlation coefficient of 95.42%in a 7-day forecast.关键词
北极海冰/人工智能/神经网络/卷积长短期记忆网络/海冰密集度Key words
Arctic sea ice/artificial intelligence/neural networks/ConvLSTM/sea ice concentration分类
天文与地球科学引用本文复制引用
夏成龙..基于ConvLSTM网络的北极海冰时空序列预测研究[J].海洋测绘,2024,44(4):48-53,6.
