ARIMA模型在河南省甲型病毒性肝炎发病数预测中的应用
Application of ARIMA model in prediction of Hepatitis A, Henan Province
高云云 1李军 2杨海燕 1陈帅印1
作者信息
- 1. 郑州大学公共卫生学院,河南郑州450001
- 2. 河南省疾病预防控制中心,河南郑州450016
- 折叠
摘要
目的 建立乘积季节自回归移动平均(ARIMA)模型,利用该模型预测河南省甲肝发病情况并探讨其可行性.方法 对2008年1月-2015年8月河南省的甲肝疫情监测数据差分平稳化,通过专家建模器筛选最优模型,利用2015年9月-2016年8月的甲肝疫情资料来拟合预期值并评价该模型的可行性.结果 2008-2015年河南省甲肝发病数逐年减少且呈现明显的季节效应;本次研究中乘积季节ARIMA(0,1,1)×(1,0,1)12模型能较好的拟合既往的甲肝报告病例数,模型统计量Ljung-Box Q为21.742,P为0.115)0.05,残差序列为白噪音;且对2015年9月-2016年8月按月报告的甲肝病例数的预测值与实际值吻合情况良好,平均误差绝对值4.67,平均相对误差绝对值为0.2.结论 ARIMA模型能较好模拟、预测河南省甲肝的发病情况,该模型的预测效能将优化甲肝预防工作,有较好的推广价值.
Abstract
Objective To establish multiple seasonal autoregressive integrated moving average (ARIMA) model for predicting the incidence of Hepatitis A and explore it's applied feasibility.Methods Difference method was conducted to smooth the data of reported cases of Hepatitis A of 2008.1-2015.8 in Henan province and build an optimal ARIMA model.Select the reported cases of Hepatitis A in 2015.9-2016.8 to assess the predictive power of the ARIMA model.Results Incidence of hepatitis A in 2008-2015 of Henan province showed the tendency of reducing year by year and the obvious seasonal effect.The sequence was a nonstationary time series.In this research,ARIMA(0,1,1) × (1,0,1)12 model fitted the incidence of previous month well,and statistics of Box-Ljung Q was insignificant which could indicate that residual sequence was white noise.Incidence from September in 2015 to August in 2016 predicted by the model was mostly consistent with the actual incidence.The mean of absolute error and the relative error were 4.67 and 0.2 respectively.Conclusion ARIMA model could successfully simulate and predict the reported cases of Hepatitis A in Henan province The Hepatitis A incidence trends predicted by this model will assist the optimization of Hepatitis A prevention,which could be applied for development and application value.
关键词
乘积季节自回归移动平均模型/甲型病毒性肝炎/疾病预测Key words
ARIMA model/Hepatitis A/Prediction引用本文复制引用
基金项目
郑州大学青年教师启动基金(32210273)
出版年
2017
NETL
NSTL
维普
万方数据