摘要
将逐步回归融入到时间序列预测模型的建立中,摒弃了传统的'考虑所有变量'模式,利用'有进有出'的形式,分清各因子主次关系,仅选用影响显著的变量建立预测方程。径向基函数人工神经网络(RBF-ANN)属于局部逼近网络,准确度高。以桦甸市五道沟站的月降水量和月蒸发量为例,分别用传统、逐步回归时间序列分析和RBF-ANN建立降水预测模型,并对比其精度。结果表明:传统、逐步回归时间序列及RBF-ANN模型的后验差比值分别为0.315、0.272、0.284,平均绝对误差分别为18.37、15.65、13.82mm,有效系数分别为0.87、0.94、0.93,精度均满足要求,最后用逐步回归时间序列法预测了未来5年的月降水量和月蒸发量。
With integration of stepwise regression into the foundation of time series analysis model,the traditional mode of ＂take into account all the variables＂ is abandoned and just significant variables are used to establish the prediction equation in the form of ＂both enter and exit＂ mode,with the distinction of each factor＇s major and minor relationship.The radial basis function artificial neural network（RBF-ANN） belongs to partial approaches network and has high accuracy.Take Huadian County＇s month precipitation as an example,and compare the accuracy of prediction equations which are established using traditional,stepwise regression time series analysis model and RBF-ANN.The results show that the posterior error ratios of the traditional time series,stepwise regression time series and RBF-ANN models are 0.315,0.272 and 0.284,the average absolute errors are 18.37 mm,15.65 mm and 13.82 mm,and the effective coefficients are 0.87,0.94 and 0.93.At last,we forecast the precipitation and evaporation in future three years with the stepwise regression time series analysis model.
出处
《重庆大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第11期131-135,共5页
Journal of Chongqing University
基金
国家自然科学基金资助项目(41072171)
关键词
时间序列
逐步回归
RBF-ANN
月降水量
预测
time series
stepwise regression
RBF-ANN
monthly total precipitation
forecasting
