神经网络BP模型用于月降水预报的研究

NEUROID BP-TYPE MODEL APPLIED TO THE STUDY OF MONTHLY RAINFALL FORECASTING

关键词:  神经网络  BP型三层映射模式  月雨量预报

Key words:  Neuroid  BP-type three-layer mapping model  Monthly rainfall forecasting

摘要:

采用神经网络BP型三层映射模式,以南京1946-1985年40年月雨量为基础序列,确定三层模式的形式i×j=8×3.k=1。通过不断调整权重系数,作出1986年1-12月的月雨量长期预报,又用同样方法但改用前一个月的实测值报后一个月的月雨量作出1986年各月的月雨址预报。平均绝对误差分别为6.07mm和5.73mm.对1994年6、7、8、9月月雨量以4个不同的起始值进行神经网络预测,都得到1994年夏季南京特旱的结果,与实测结果相同。

Abstract:

A neuroid BP-type three-layer mapping model is used for monthly rainfall forecasting with 1946－1985 Nanjing monthly precipitation records as basic sequences and the model has the form i×j=8×3.k=1 By steadily modifying the weighing coefficient, long-range monthly forecasts for June through December 1986 are constructed and month-to-month predictions are made based on the records of the preceeding months of the year, with mean absolute error reaching 6.07 and 5. 73mm, respectively. Additionally, with a set of four monthly initial valuesi; June through September, 1994, neuroid forecasting is indicating the same drought result in Nanjing durulg the summer, an outcome that is in good agreement with the observation.