基于最优线性组合方法的甲型病毒性肝炎发病数预测

Prediction of the incidence of hepatitis A with optimal linear composite prediction model

目的：利用SARIMA、RBFNN、组合模型对2011年甲型病毒性肝炎月发病数进行预测并比较,探讨最优线性组合模型在甲型病毒性肝炎发病数预测中的实用价值。方法选取2005年至2010年我国内地法定报告的甲型病毒性肝炎月发病数资料分别建立季节性求和自回归移动平均模型以及径向基神经网络模型后,将2两种模型的拟合结果与原始发病数建立线性回归模型,比较3种模型的预测精度。结果 SARIMA模型预测结果的平均绝对误差、平均相对误差以及均方误差值分别是413.667、0.154、0.392。径向基神经网络预测结果的平均绝对误差、平均相对误差以及均方误差值分别是291.833、0.118、0.344。组合模型拟合预测结果的平均绝对误差,平均相对误差以及均方误差值分别是202.333、0.082、0.286。由此得出,组合模型的拟合值及其预测值的3个指标均最小,其次为径向基神经网络,SARIMA的各指标值最大。结论最优线性组合模型在甲型病毒性肝炎发病数预测中具有较高的预测精度,可以用于指导甲型病毒性肝炎的预防与控制。

Objective To explore the application value of optimal linear composite prediction model in forecasting the incidence number of hepatitis A （ HA） . Methods Seasonal autoregressive integrated moving average （ SARIMA） model and ra-dial basis function neural network （ RBFNN） were developed based on the monthly incidence of HA in China's Mainland from 2005 to 2010. Linear regression model between the true incidence and the simulated values of SARIMA and RBFNN were also devel-oped. The incidence values of 2011 were predicted with the three models and their forecasting efficiency was compared. Results The mean average error （ MAE） , mean average percentage error （ MAPE） and root mean square error （ RMSE） of predicted values by SARIMA model were 413. 667, 0. 154, 0. 392 respectively. The MAE, MAPE, RMSE of predicted values by RBFNN were 291. 833,0. 118,0. 344 respectively. The MAE, MAPE, RMSE of predicted values by optimal linear composite prediction model were 202. 333,0. 082,0. 286 respectively. So the MAE, MAPE, RMSE of predicted values by optimal linear composite prediction model were all lower than those individual models. Conclusion The result indicates that the optimal linear composite prediction model can be well applied to forecast the incidence of HA.