3种模型在肺结核发病率预测中的比较研究

目的对GM（1,1）模型、霍尔特双参数指数平滑预测模型和ARIMA模型在肺结核发病率预测中的效果进行比较。方法利用1980-2007年北京市肺结核的发病率分别建立GM（1,1）灰色预测模型、霍尔特双参数指数平滑预测模型和ARIMA模型,对建立的模型进行拟合。比较3个模型的拟合效果,同时利用ARIMA模型对2008年北京市的肺结核发病率进行预测。结果针对北京市肺结核发病率建立的GM（1,1）模型、霍尔特双参数指数平滑预测模型和ARIMA模型的平均误差率（MER）分别为15.11%、9.51%、9.52%,决定系数R2分别为0.935、0.964、0.969。结论 ARIMA模型对于隐含波动周期并且不稳定的循环型时间序列拟合效果优于GM（1,1）模型,对解决时间序列类型的肺结核发病率等资料有很好的实用价值。

Comparison of three models on prediction of incidence of pulmonary tuberculosis in Beijing

Objective To compare the effects of GM（1,1） model,Holt＇s linear exponential smoothing model and ARIMA model on prediction of incidence of pulmonary tuberculosis.Methods GM（1,1） model、Holt＇s linear exponential smoothing model and ARIMA model were established with the pulmonary tuberculosis incidence of Beijing from 1980 to 2007 and the three models were fit together,and the effects of fitting were compared.Use ARIMA model to predict the pulmonary tuberculosis incidence in Beijing in 2008.Results The MER of GM（1,1）model、Holt＇ s linear exponential smoothing model and ARIMA model for Beijing were 15.11%、9.51%、9.52% respectively and the R2 of the three models were 0.935、0.964 and 0.969.Conclusion ARIMA model is a good choice in unstable and recycling time series data on fitting and forecasting,so it is of practical value in dealing with time series data such as the incidence of pulmonary tuberculosis.