摘要
目的:探讨时间序列ARIMA模型在时间序列资料分析中的应用,建立咳嗽症状监测数据的预测模型.方法:采用条件最小二乘方法估计模型参数.通过对数转换及差分方法使原始序列平稳,按照残差不相关原则、简洁原则确定模型结构,依据AIC和SBC准则确定模型阶数,最终建立起ARIMA预测模型.结果:ARIMA(1,1,1)模型拟合效果较好,方差估计值为0.7361,AIC=95.6092,SBC=98.8310,对模型进行白噪声残差检验,提示残差为白噪声.结论:症状监测这种具有时间序列特点的资料可以用ARIMA模型来进行拟合估计.本文中预测结果可信区间比较宽,可能是因为时间序列比较短,还未能考虑到季节趋势.另外,所用监测数据是在中小学生在校发生症状的人数,故在节假日会出现缺失值,样本量和时间长度均有限,可能影响模型估计的准确性,本研究的结论还有待于将来资料积累后进行修正和深化.
Objective To discuss the application of ARIMA model on data of time series and fit predictive model on syndromic surveillance.Methods Parameter of model was estimated based on conditional least squares.The structure was determined according to criteria of residual un-correlation and concision.ARIMA predictive model was fitted and the order of model was confirmed through Akaiake Information Criterion and Schwarz Bayesian Criterion.Results The effect of ARIMA（1,1,1） model was better than others.The estimation of variance was 0.7361, AIC=95.6092,SBC=98.8310.The analysis of white-noise residual of model showed that residual was white-noise series.Conclusion ARIMA model can be suitably applied on data of time series of symdromic surveillance.The credibility interval of forecast was a little wide,which might due to the time series were relatively short and the process of modeling didn＇t take the seasonal trends into account.In addition,there were missing values because of holidays.The accuracy of model may be affected by the relatively small sample size and the length of time series.The conclusion of this study has yet to be further confirmed in future studies.
出处
《生物数学学报》
CSCD
北大核心
2011年第3期563-568,共6页
Journal of Biomathematics
基金
2009年浙江省医药卫生科学研究基金计划(A类)(2009A175)
