摘要
目的比较求和自回归滑动平均(ARIMA)模型、ARIMA与广义回归神经网络(GRNN-ARIMA)组合模型、径向基函数(RBF)网络模型在肺结核发病预测中的应用,探讨优化模型,为完善结核病预测预警系统提供建议和资料。方法利用1998年1月-2011年6月全国肺结核逐月发病资料构建ARIMA模型,而2011年7-12月数据作为模型测试值;将上述ARIMA模型拟合值作为GRNN模型输入值,各月实际发病率作为输出值,构建GRNN-ARIMA组合模型并预测;将1998年1月-2011年6月数据分段,构建三维输入,一维输出的RBF网络模型并预测。比较三种模型的拟合及预测效果优劣。结果三种模模型拟合肺结核发病情况的均方误差MSE值依次为GRNN-ARIMA(0.0848)<RBF(0.1987)<ARIMA(0.2800);三种模型预测2011年7-12月各月的发病率与实际值比较的均方误差MSE值依次为:GRNN-ARIMA(0.0571)<RBF(0.1024)<ARIMA(0.1053),其他模型评价指标也显示GRNN-ARIMA组合模型误差最小。结论 GRNN-ARIMA组合模型拟合及预测效果均优于RBF网络模型和单纯ARIMA模型,它能显著提高预测精度,具有很好的实用价值。
Objective To compare the autoregressive integrated moving average(ARIMA) model,ARIMA-generalized regression neural netw ork(GRNN) combination model and the radial basis function(RBF) network model on prediction of incidence of pulmonary tuberculosis and explore the optimized model,so as to provide suggestions and materials for improving the tuberculosis forecast and early w arning system.Methods The ARIMA model was established under the national data of the monthly incidence of pulmonary tuberculosis from Jan.1998 to Jun.2011,w hile the data from Jul.2011 to Dec.2011 w as used as testing data.The fitting values of the ARIMA model w ere used as the input of the GRNN model,and those actual values w ere used as the output,then the GRNN-ARIMA combination model w as developed.Afterw ards,w e striped the data from Jan.1998 to Jun.2011 and constructed the RBF netw ork model w ith a structure of three dimensional inputs and one dimensional output.Finally,w e compared the fitness and forecasting precision of the three models.Results The mean square error of the model fitness w as 0.0848,0.1987 and 0.2800 for GRNN-ARIMA combination model,RBF netw ork model and ARIMA model,respectively.The mean square error of the prediction accuracy w as 0.0571,0.1024,and 0.1053 for GRNN-ARIMA combination model,RBF netw ork model and ARIMA model,respectively.The results of other model evaluation indices also suggested that GRNN-ARIMA combination model had the minimum error.Conclusion According to the model fitness and prediction accuracy,the GRNN-ARIMA model is superior to RBF netw ork model and ARIMA model,having a good practical value for significantly improving the forecasting precision.
出处
《中国卫生统计》
CSCD
北大核心
2013年第4期480-483,共4页
Chinese Journal of Health Statistics
基金
国家"十一五"科技重大专项(2008ZX10003-008)
