ARIMA季节模型在预测新疆地区丙型肝炎发病数中的应用

Application of the multiple seasonal ARIMA model in predicting incidence of HCV in Xinjiang

目的研究建立新疆地区丙肝月发病数的ARIMA乘积季节模型并做短期预测,为丙肝的防控提供决策和定量参考。方法 2005年1月-2014年6月为训练集,2014年7月-2014年12月为验证集,对训练集做BoxCox变换,用非季节差分和季节差分得到平稳序列后做纯随机数检验,用ACF和PACF确定ARIMA(p,d,q)×(P,D,Q)S的参数范围,估计模型的参数并做模型和参数的显著性检验,选用AICc作为最小信息准则函数,选取最优模型并做验证集测试和预测。结果研究选用ARIMA(2,1,0)×(1,1,0)12模型,其训练集MAPE=1.44%,验证集MAPE=4.80%,平均误差2.37%,拟合效果较好。使用模型预测2015年丙肝月发病数,结果显示2015年发病数比2014年稍高,峰值为1 154例,出现在3月。结论 ARIMA(2,1,0)×(1,1,0)12模型能够较好地预测新疆地区丙肝月发病数,可为政府相关部门提供可靠信息。

Objective To explore decision and quantitative reference of the prevention and control of HVC and to make a short-term prediction,the ARIMA model was established to fitted the monthly data in Xinjiang.Methods January2005to June2014was as the training set,while July2014to December2014was as the validation set,and the training set was transformed by Box-Cox and seasonal and non-seasonal difference,and the stationary sequence was applied pure randomness test to estimate parameters of model ARIMA(p,d,q)×(P,D,Q)-S with ACF and PACF.AICc was selected as the information criterion function,testing optimal model in validation set and to make the prediction.Results The fitting effect of model ARIMA(2,1,0)×(1,1,0)12is better than others.Training set MAPE is1.44%,validation set MAPE is4.80%,and average error is2.37%.Using the model to forecast the number of incidence of HCV in2015monthly,the result shows that the incidence in2015was slightly higher than that in2014,with a peak of1154cases occurrence in March.Conclusion The model of ARIMA(2,1,0)×(1,1,0)12can be used to simulate and forecast the monthly incidence of HCV in Xinjiang,which may provide the reliable information of prevention and control for the relevant departments of government.