我国包虫病报告病例数自回归移动平均模型预测研究  

Study on the ARIMA model application to predict echinococcosis cases in China

作　　者：谭恩丽[1] 王正峰[2] 周文策[2] 李石柱[3] 卢艳[3] 艾琳[3] 蔡玉春[3] 滕雪娇 张顺先[3] 党志胜[3] 杨春利 陈家旭[3] 胡薇[3,5] 周晓农 田利光[3] 

机构地区：[1]兰州大学第一医院老年呼吸科,兰州730000 [2]兰州大学第一医院普外二科 [3]中国疾病预防控制中心寄生虫病预防控制所、国家卫生和计划生育委员会寄生虫病原与媒介生物学重点实验室、世界卫生组织热带病合作中心、国家级热带病国际联合研究中心 [4]上海市皮肤病医院 [5]复旦大学生命科学学院微生物学与微生物工程系

出　　处：《中国血吸虫病防治杂志》2018年第1期47-53,共7页Chinese Journal of Schistosomiasis Control

基　　金：国家自然科学基金(81473022)

摘　　要：目的采用自回归移动平均模型(Autoregressive integrated moving average,ARIMA)对全国(不含港、澳、台地区)包虫病月报告病例数进行预测,为包虫病的防控提供科学参考。方法通过SPSS 24.0软件,分别以2007-2015年和2007-2014年全国包虫病月报告病例数,分别建立最优的ARIMA模型,并进行模型比较。结果 2007-2015年全国包虫病月报告病例数的最优模型为ARIMA(1,0,0)(1,1,0)_(12),预测相对误差为-13.97%,AR(1)=0.367(t=3.816,P<0.001)、SAR(1)=-0.328(t=-3.361,P=0.001),Ljung-Box Q=14.119(df=16,P=0.590)。2007-2014年全国包虫病月报告病例数的最优模型为ARIMA(1,0,0)(1,0,1)12,预测相对误差为0.56%,AR(1)=0.413(t=4.244,P<0.001),SAR(1)=0.809(t=9.584,P<0.001),SMA(1)=0.356(t=2.278,P=0.025),Ljung-Box Q=18.924(df=15,P=0.217)。结论时间序列不同,所建立的预测模型可能不同。数据积累越多、预测时间越短、预测误差越小的情况还需得到进一步验证。模型的建立和预测应用是动态过程,需要不断根据积累的数据进行调整,但同时要充分考虑影响传染病报告病例数相关工作(普查和专项调查等)的影响。Objective To predict the monthly reported echinococcosis cases in China with the autoregressive integrated mov-ing average（ARIMA）model,so as to provide a reference for prevention and control of echinococcosis. Methods SPSS 24.0 software was used to construct the ARIMA models based on the monthly reported echinococcosis cases of time series from 2007 to 2015 and 2007 to 2014,respectively,and the accuracies of the two ARIMA models were compared. Results The model based on the data of the monthly reported cases of echinococcosis in China from 2007 to 2015 was ARIMA（1,0,0）（1,1, 0）12,the relative error among reported cases and predicted cases was-13.97%,AR（1）=0.367（t=3.816,P〈0.001）,SAR （1）=-0.328（t=-3.361,P=0.001）,and Ljung-Box Q=14.119（df=16,P=0.590）.The model based on the data of the monthly reported cases of echinococcosis in China from 2007 to 2014 was ARIMA（1,0,0）（1,0,1）12,the relative error among reported cases and predicted cases was 0.56%,AR（1）=0.413（t=4.244,P〈0.001）,SAR（1）=0.809（t=9.584, P〈0.001）,SMA（1）=0.356（t=2.278,P=0.025）,and Ljung-Box Q=18.924（df=15,P=0.217）.Conclusions The different time series may have different ARIMA models as for the same infectious diseases.It is needed to be further verified that the more data are accumulated,the shorter time of predication is,and the smaller the average of the relative error is.The estab-lishment and prediction of an ARIMA model is a dynamic process that needs to be adjusted and optimized continuously accord-ing to the accumulated data,meantime,we should give full consideration to the intensity of the work related to infectious diseas-es reported（such as disease census and special investigation）.

关 键 词：包虫病 月报告病例数 自回归移动平均模型 建模 

分 类 号：R53[医药卫生—内科学]