|国家科技期刊平台
首页|期刊导航|Atmospheric and Oceanic Science Letters|Comparative study on the performance of ConvLSTM and ConvGRU in classification problems-taking early warning of short-duration heavy rainfall as an example

Comparative study on the performance of ConvLSTM and ConvGRU in classification problems-taking early warning of short-duration heavy rainfall as an exampleOA

中文摘要

卷积长短期记忆单元ConvLSTM和卷积门控循环单元ConvGRU是两种广泛应用的深度学习单元,通过将循环机制与卷积运算相结合,常常用于时空序列的预测.为了明确上述两种模型的收敛速度和分类能力,需要使用相同的模型架构对相同的分类问题进行预测.本研究将北京短时强降水区级预警问题看作深度学习中的二分类问题,使用京津冀雷达网的组合反射率数据和北京区域内的自动气象站降雨数据进行深度学习模型的训练和评估.结果表明,ConvGRU的收敛速度比ConvLSTM快约25%.ConvLSTM和ConvGRU的预警性能随地区,时间,降雨强度的变化趋势相似,但大部分ConvLSTM的得分较高,少数情况下ConvGRU的得分较高.

Meng Zhou;Jingya Wu;Mingxuan Chen;Lei Han;

Faculty of Information Science and Engineering,Ocean University of China,Qingdao,China Institute of Urban Meteorology,China Meteorological Administration,Beijing,ChinaInstitute of Urban Meteorology,China Meteorological Administration,Beijing,ChinaFaculty of Information Science and Engineering,Ocean University of China,Qingdao,China

大气科学

深度学习卷积长短期记忆单元卷积门控循环单元分类问题

《Atmospheric and Oceanic Science Letters》 2024 (004)

P.52-57 / 6

This study was jointly funded by the National Key R&D Program of China[grant number 2022YFC3004103];the National Natural Foundation of China[grant number 42275003];the Beijing Science and Technology Program[grant number Z221100005222012];the Beijing Meteorological Service Science and Technology Program[grant number BMBKJ202302004];the China Meteorological Administration Youth Innovation Team[grant number CMA2023QN10].

10.1016/j.aosl.2024.100494

评论