基于主成分分析和LSTM神经网络的海温预报模型  

SST forecasting model based on principal component analysis and LSTM neural network

作　　者：李竞时 匡晓迪[1,2] 李琼 何恩业[1,2] 张聿柏 袁承仪 张延琳 LI Jingshi;KUANG Xiaodi;LI Qiong;HE Enye;ZHANG Yubai;YUAN Chengyi;ZHANG Yanlin(National Marine Environmental Forecasting Center,Beijing 100086,China;Key Laboratory of Marine Hazards Forecasting,National Marine Environmental Forecasting Center,Ministry of Natural Resources,Beijing 100081,China;Shandong Marine Forecast and Hazard Mitigation Service,Qingdao 266104,China;Tianjin University of Science and Technology,Tianjin 300222,China;Liaoning Natural Resources Affairs Service Center,Shenyang 110033,China)

机构地区：[1]国家海洋环境预报中心,北京100081 [2]国家海洋环境预报中心自然资源部海洋灾害预报技术重点实验室,北京100081 [3]山东省海洋预报减灾中心,青岛266104 [4]天津科技大学,天津300222 [5]辽宁省自然资源事务服务中心,辽宁沈阳110033

出　　处：《海洋预报》2023年第2期1-10,共10页Marine Forecasts

基　　金：自然资源部海洋环境信息保障技术重点实验室开放基金课题资助;海洋预警监测(海温预报释用服务试点)(SDGP3700000002021-02003589);国家自然科学基金(41606028)。

摘　　要：利用荣成、海阳两站的自建浮标海温观测数据以及区域大气模式WRF(Weather Research and Forecasting)的气象数值预报数据,基于主成分分析(Principal Component Analysis,PCA)法和长短时记忆(Long Short-Term Memory,LSTM)神经网络,提出了适用于单站海表温度预报的PCALSTM海温预报模型。该模型可以提供24~120 h预报时效的海温预报,预测效果比数值模型和统计模型明显提高。Using the sea temperature observation data of buoys at Rongcheng and Haiyang marine stations and the numerical forecast meteorology data of the regional atmospheric model Weather Research and Forecasting(WRF),and based on the Principal Component Analysis(PCA)and Long Short-Term Memory(LSTM)neural network,a PCA-LSTM sea temperature forecasting model suitable for the Sea Surface Temperature(SST)forecasting is proposed in this paper.This model can provide SST forecast for the following 24~120 hours,and its forecasting accuracy is significantly improved compared with the numerical model and statistical model.

关 键 词：主成分分析 长短时记忆神经网络 海温预报 

分 类 号：P731.31[天文地球—海洋科学]