基于长短记忆神经网络的降水预报订正方法研究  

Research on the correction method of precipitation forecast based on long short memory neural network

作　　者：任鸿飞 闫霜 周一诺 REN Hongfei;YAN Shuang;ZHOU Yinuo(Miyun District Meteorological Bureau,Beijing 101500,China)

机构地区：[1]北京市密云区气象局,北京101500

出　　处：《中国高新科技》2024年第10期36-38,共3页

基　　金：北京市气象局科技项目青年基金(BMBKJ202201017)。

摘　　要：数值预报系统是降水预报的现行方法,然而由于初值和数值预报自身的不确定性,使得降水预报存在一定的误差。文章基于自动气象站观测实况数据和数值预报系统预报输出的降水预报产品。通过长短记忆神经网络(LSTM)算法,建立和训练降水订正模型。实验证明,该订正方法对数值预报系统降雨预报值小雨量级结果订正具有重要意义,从而改善降水预报准确性,更好地为气象预报业务提供有力技术支撑。Numerical forecasting systems are the current method for precipitation forecasting,but due to the uncertainty of initial values and numerical forecasting itself,there are certain errors in precipitation forecasting.The article is based on the precipitation forecast products of automatic weather station observation data and numerical forecast system output.Establish and train a precipitation correction model using Long Short-Tern Memory(LSTM)algorithm.Experimental results show that this correction method is of great significance for the correction of small rainfall magnitude results in numerical forecasting systems,thereby improving the accuracy of precipitation forecasting and providing strong technical support for meteorological forecasting operations.

关 键 词：神经网络 降水订正 数值预报系统 

分 类 号：P457[天文地球—大气科学及气象学]