基于ARIMA模型的流感样病例预警预测分析

目的探索建立适合于流感样病例预测的自回归求和移动平均模型（ARIMA模型）。方法采集深圳市南山区2006--2011年流感样病例监测数据，绘制序列图，差分使序列平稳化，通过自相关分析和偏相关分析进行模型识别，根据AIC（赤池信息准则）和BIC（贝叶斯信息准则）确定模型参数，建立ARIMA预测模型，用Q统计量法对模型适用性进行检验，用2012年全年实际监测数据与模型预测值进行比较，评价模型预测效果。结果2006--2011年流感样病例累计报告199360例，月发病最大值9765例，月发病最小值594例，平均月发病2769例。通过对2006--2011年各月的监测数据进行分析发现，各年度流感样病例发病呈现明显的高峰和低谷，高峰在每年5—8月份，低谷在当年的11月份至次年2月份，不同年度略有波动。对序列进行一阶差分后可得到较为平稳的序列，适合进行模型拟合，经过模型拟合诊断发现ARIMA（0，1，1）×（0，0，1）12模型为最优模型，AIC值和BIC值最小，分别为1239．19和1245．98，Box-Ljung检验结果Q值为19．07，P〉0．05，通过2012年拟合值与实际值比较，结果差异无统计学意义（P〉0．05）。结论ARIMA模型可以较好地对流感样病例进行拟合分析预测。

Prediction of influenza like illness incidence based on ARIMA model

Objective To build appropriate prediction model of influenza like illness （ILI） using Autoregressive Integrated Moving Average （ARIMA） model. Methods We collected the data of ILl surveillance from 2006 to 2011 in Nanshan District, Shenzhen, and built ARIMA model according to Akaike＇ s Information Criterion （AIC） and Bayesian Information Criterion （BIC）. The autocorrelation analysis and partial correlation analysis were used to identify the model. The model diagnosis was performed using Q statistic analysis. The actual ILI surveillance data in 2012 were compared with predictive value of the model to evaluate its predictive effect. Results A total of 199 360 ILl cases were reported from 2006 to 2011. The month max was 9 765 cases, the month rain was 594 cases, and the average was 2 769 cases per month. The annual incidence of ILI cases presented obvious peaks and valleys in 2006 -2011. The incidence peak was from May to August and the incidence valley was from November to February each year. Relatively smooth sequence was obtained and suitable for model fitting. ARIMA （0,1,1）× （0,0,1）12 was selected as the optimal model. AIC and BIC values were the least, 1 239. 19 and 1 245.98, respectively. The Q statistic was 19.07 （P 〉 0.05 ） by Box-Ljung testing, indicating the applicability of the model. There was no statistically significant difference between the observed value in 2012 and predicted value （P 〉 0. 05 ）. Conclusion ARIMA model is suitable for prediction of ILI incidence.