基于深度学习的大风订正预报研究OA北大核心CSTPCD
Research on wind speed forecasting correction based on deep learning
基于数值预报模式产品的风速预报集成学习误差订正方法,通过长短期记忆网络(LSTM)和残差神经网络(ResNet)构建新的风速预测混合模型ResNet-LSTM.采用2019-2020年欧洲中期天气预报中心39种数值天气预报模式产品训练深度学习模型,对格点预报产品插值到站点后的预报结果进行订正.结果表明:与ECMWF的原始预报相比,ResNet-LSTM模型在预测6级以上阵风时的TS评分整体可以提高50％以上,预报精度提升.寒潮大风和台风大风的个例分析也表明,ResNet-LSTM可以有效解决大风漏报问题,对站点风速的预报订正效果显著.
A novel wind speed prediction model,the ResNet-LSTM model,is proposed combining the Long Short-Term Memory(LSTM)model and Residual Network(ResNet)model.By using 39 kinds of numerical weather forecasting products from the European Center for Medium Range Weather Forecasting(ECMWF),a deep learning model is trained to correct wind speed forecasts.The results show that compared with the ECMWF results,the TS score of the ResNet-LSTM model for gusts above level 6 has been increased by over 50％.Further analysis shows that the ResNet-LSTM model can effectively solve the fail report problem and improve wind speed forecasting corrections.
杨凡;刘志丰;任兆鹏;崔天伦;于洋
青岛市气象服务中心,山东青岛 266003青岛市黄岛区气象局,山东青岛 266400青岛市气象服务中心,山东青岛 266003青岛天洋气象科技有限公司,山东青岛 266400青岛天洋气象科技有限公司,山东青岛 266400
大气科学
残差神经网络长短期记忆网络风速预报订正
ResNet modelLong Short Term Memory neural network,wind speedforecasting,correction
《海洋预报》 2024 (6)
23-31,9
青岛市气象局课题(2021qdqxz02、2019qdqxz01)山东省气象局项目(2022SDQN06).
评论