基于ARIMA-GRNN组合模型预测我国甲肝发病率

目的 探讨ARIMA-GRNN组合模型在甲肝发病序列预测中的应用，并比较其与ARIMA模型和BPNN模型的预测效果。方法 通过收集2004年1月～2014年12月我国甲肝发病序列资料，用SPSS 13.0建立ARIMA模型，用Matlab 8.0建立BPNN模型和ARIMA-GRNN组合模型，并用2014年数据对模型的预测效果进行评价。结果 针对我国甲肝发病序列建立的三种预测模型拟合的平均相对误差的值依次为:ARIMA模型（7.29 785）

Forecasting incidence of hepatitis A with ARIMA-GRNN hybrid model in China

Objective To explore the application of ARIMA-GRNN hybrid model, ARIMA model and BPNN model in the prediction on incidence of hepatitis A in China and compare the predicative effect among them. Methods The data of incidence of hepatitis A from January 2004 to December 2014 in China was collected and SPSS 13.0 was used to construct ARIMA model, Matlab 8.0 was used to establish BPNN model and ARIMA-GRNN hybrid model, afterwards the data in 2014 was used to evaluate the effect of prediction. Results The average relative error of the three models, which was established to predict the incidence of hepatitis A sequence in China, fitness was 7.29785, 5.86174 and 4.91063 for ARIMA model, BP network model and ARIMA-GRNN hybrid model, respectively. The average relative error of accuracy predicted by ARIMA model, BP network model and ARIMA-GRNN hybrid model in 2014 was 6.44 067, 5.74 400 and 4.86 292, respectively. The results of other model evaluation indices, such as average error rate, mean square predict error and mean absolute error, also showed that ARIMA-GRNN hybrid model had the minimum error. Conclusions ARIMA-GRNN combination model is superior to the ARIMA model and BPNN model.