中文摘要：

【提要】目的阐述ARIMA—BPNN组合模型预测流感发病率的方法和步骤，探讨其在流感发病率NN中的应用。方法利用河南省2004年1月-2010年12月的流感疫情数据作为训练集，建立ARIMA模型和ARIMA—BPNN组合模型，选取2011年1月-12月的疫情数据作为检验集，评价模型的预测效能。结果ARIMA（3，0，0）模型和ARIMA—BPNN组合模型预测值的平均绝对误差及平均误差绝对率分别为1．438、27．65％和0．029、0．43％，ARIMA—BPNN组合模型的预测效能优于ARIMA模型。结论ARIMA—BPNN组合模型能有效模拟、预测流感的发病疫情，具有较好的推广应用价值。

英文摘要：

Objective To describe the procedure of building ARIMA model and ARIMA-BPNN hybrid model, and ex- plore the value of potential application of the above model to predict the incidence of influenza. Methods According to the au- toregressive integrated moving average and back-propagation neural network theory, ARIMA and ARIMA-BPNN hybrid models were established using the epidemic data of influenza in Henan province from January 1,2004 to December 31,2010, and the predictive performances were measured and evaluated using the data from January 1 to December 31,2011. Results The mean absolute error（MAE） and mean error rate（MER） of the optimal ARIMA model and ARIMA-BPNN hybrid model were 1. 438, 27.65 % and 0. 029,0.43 %, respectively, which suggested that the ARIMA-BPNN hybrid model＇s predictive efficacy was better than the ARIMA model. Conclusion ARIMA-BPNN hybrid model could effectively fit and predict the incidence of influenza, which was of great application value for the prevention and control of influenza.