两种时间序列分析模型在甲肝预测中的应用比较  

Comparison of applying two time series models in the predicting of hepatitis A

作　　者：汪鹏[1] 彭颖[1] 张靖[1] 金小毛[1] 陈邦华[1] Wang Peng;Peng Ying;Zhang Jing;Jin Xiaomao;Chen Banghua(Public Health Emergency Management Office,Wuhan Center for Disease Control and Prevention,Wuhan 430024,China)

机构地区：[1]武汉市疾病预防控制中心应急办,430024

出　　处：《国际病毒学杂志》2022年第2期125-128,共4页International Journal of Virology

基　　金：武汉市卫生健康科研基金(WG19Q05)。

摘　　要：目的比较差分自回归移动平均模型和指数平滑模型在全国甲肝发病情况预测中的效果,为甲肝监测预警提供合适的数学模型。方法从公共卫生科学数据中心收集2010—2020年全国甲肝逐月发病数据,分别用差分自回归移动平均模型和指数平滑模型进行拟合,筛选出最优的差分自回归移动平均模型和指数平滑模型,再用最优模型分别预测2021年1—10月全国甲肝发病数,并比较预测精度。结果ARIMA(1,1,0)(1,0,0)12是最优的差分自回归移动平均模型,Holt-winters乘法模型是最优的指数平滑模型,两种模型预测的平均绝对百分比误差分别为15.64%和13.08%,平均绝对误差分别为144和124。结论Holt-winters乘法模型在全国甲肝逐月发病数预测中的精度更高,可用于数据波动不大时间序列的拟合预测。Objective To compare the efficiency of autoregressive integrated moving average(ARIMA)model and exponential smoothing model in predicting the reported cases of hepatitis A in China,so as to provide an appropriate model for the surveillance and early warning of hepatitis A.Methods Monthly reported cases of hepatitis A in China between 2010 to 2020 were collected from the Data-center of China Public Health Science.ARIMA model and exponential smoothing model were used for data fitting to get the optimal models,respectively.The selected models were used to predict the monthly reported cases from Jan to Oct 2021 and the prediction accuracies were compared.Results The optimal ARIMA model was ARIMA(1,1,0)(1,0,0)12 while Holt-Winters multiplicative model was the optimal model for exponential smoothing method.For the two models,the mean absolute percentage errors were 15.64%and 13.08%,and the mean absolute errors were 144 and 124,respectively.Conclusions Holt-Winters multiplicative model showed better prediction accuracy,and can be used in fittings and predictions in time series with minor fluctuation.

关 键 词：ARIMA 指数平滑 时间序列 甲肝 预测 

分 类 号：R512.61[医药卫生—内科学] R181.3[医药卫生—临床医学]