基于LSTM深度神经网络的精细化气温预报初探

FINE TEMPERATURE FORECAST BASED ON LSTM DEEP NEURAL NETWORK

利用LSTM(Long Short-Term Memory)深度神经网络和空军T511数值预报产品,对宝鸡市2017年9月到2018年3月每日逐3小时实况观测的数据进行模拟分析,建立宝鸡市未来24小时精细化气温预报模式。结果表明:其精细化气温预报准确率为68. 75%,日最低气温预报准确率为84. 62%,日最高气温预报准确率为61. 54%,并能较好地对天气过程转折进行刻画,可满足日常气温预报的需要。

LSTM deep neural network and Air Force T511 numerical forecast product were used to simulate and analyze the daily 3-hour real-time observation data of Baoji city from September 2017 to March 2018.And a fine temperature forecast model for Baoji in the next 24 hours was established.The results show that the accuracy of fine temperature forecast is 68.75%,the accuracy of minimum daily air temperature forecast is 84.62%,and the accuracy of maximum daily air temperature forecast is 61.54%.And the model can better describe the weather transition,which can meet the needs of daily temperature forecast.