神经网络方法在广西日降水预报中的应用

Application of Artificial Neural Network to Daily Regional Average Rainfall Forecast in Guangxi

以广西前汛期5、6月区域平均日降水量作为预报对象,采用人工神经网络方法进行新的数值预报产品释用预报研究。对T213预报因子进行自然正交分解,有效浓缩数值预报产品因子的预报信息,并结合日本降水预报模式因子建立广西3个不同区域的逐日降水神经网络释用预报模型。运用与实际业务预报相同的方法对2004年5、6月进行逐日的实际预报试验,并与T213的降水预报进行对比分析。结果表明,本文建立的3个区域日平均降水量神经网络预报模型,在预报性能上明显优于同期的T213降水预报。

A new forecast application of numerical forecast products is exemplified by use of artificial neural network （ANN） to daily forecasts of the regional average precipitation of Guangxi in May and June. Three different regional rainfall forecast models are established based on the useful information concentrated from numerous T213 factors by Empirical Orthogonal Function （EOF）, and the forecast factors of the Japanese precipitation model. Daily operational forecast trial of the three regional ANN models were made in May and June 2004 ,and their forecasts compared day-by-day with ones of T213 numerical model. The results show that the three regional ANN forecast models are superior to T213 in daily precipitation forecast in the period. The prospects for the application of numerical forecast products are encouraging.