相似偏差订正法在短期温度预报中的应用研究

Application of similar deviation correction method in short-term temperature forecast

为提高温度预报精度,本文提出一种新的相似偏差订正法建立短期温度预报模型,并与气象业务常用的多元回归法、BP神经网络法进行对比。结果表明:(1)温度预报精度均具有明显日变化特征,午后精度较高,而凌晨精度偏低;(2)基于20:00起报资料得到的温度预报精度略高于08:00起报资料;(3)温度预报精度由高到低的顺序依次为相似偏差订正法、BP神经网络法、多元回归法和ECMWF模式产品的2 m温度,若从制作短期逐时温度预报的精度、合理性及运行效率等方面考虑,相似偏差订正法优于BP神经网络法和多元回归法。

In order to improve the accuracy of temperature forecast,a new similar deviation correction method was proposed to establish a short-term temperature forecast model,and was compared with multiple regression method and BP neural network method that commonly used in meteorological operations.The results show that(1)the accuracy of temperature prediction has obvious diurnal variation,and the accuracy in the afternoon is higher than that in the morning.(2)The accuracy of temperature forecast based on the data from 8 pm is slightly higher than that from 8 am.(3)The order of temperature forecast accuracy from high to low is similar deviation correction method,BP neural network method,multiple regression method,and 2 m temperature of ECMWF model products.If the accuracy,rationality and operation efficiency of short-term hourly temperature forecast are considered,similar deviation correction method is better than BP neural network method and multiple regression method.