SARIMA-GRNN组合模型在伤寒与副伤寒逐月发病数预测中的应用

  目的  建立季节性差分自回归移动平均(seasonal autoregressive integrated moving average, SARIMA)-广义回归神经网络(generalized regression neural network, GRNN)组合模型，为伤寒与副伤寒发病数的预测提供方法学上的新思路。  方法  利用2011年1月-2019年12月中国伤寒与副伤寒逐月发病数资料，分别构建SARIMA模型和SARIMA-GRNN组合模型，比较两种模型的拟合和预测效果。  结果  最优的SARIMA模型为SARIMA (2, 1, 1) (0, 1, 1)₁₂，SARIMA-GRNN组合模型的最优光滑因子(spread)为0.21。评价SARIMA-GRNN组合模型拟合效果的参数均方根误差(root mean squared error, RMSE)、平均绝对误差(mean absolute error, MAE)和平均绝对百分比误差(mean absolute percentage error, MAPE)为90.08、71.44和7.07%，分别小于SARIMA模型的99.44、79.15和7.86%；评价预测效果的RMSE、MAE和MAPE为100.86、75.94和9.57%，均小于SARIMA模型的125.44、97.33和10.89%。  结论  SARIMA-GRNN组合模型比传统SARIMA模型更能拟合中国伤寒与副伤寒逐月的发病数，而且预测精度更高，可应用于伤寒与副伤寒逐月发病数的预测。

Application of SARIMA-GRNN combined model in forecasting the monthly incidence of typhoid fever and paratyphoid fever

  Objective  This study aimed to establish a seasonal autoregressive integrated moving average (SARIMA)-general regression neural network (GRNN) combined model, so as to provide new methodological ideas for forecasting the incidence of typhoid fever and paratyphoid fever.  Methods  Using data of typhoid fever and paratyphoid fever from January 2011 to December 2019, the SARIMA model and the SARIMA-GRNN combined model were constructed respectively, and the fitting and forecasting effects of the two models were compared.  Results  The optimal SARIMA model was SARIMA (2, 1, 1) (0, 1, 1)₁₂ and the optimal smoothing factor of SARIMA-GRNN combined model was 0.21. The root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) of the SARIMA-GRNN combined model fitting effect were 90.08, 71.44, and 7.07%, which were smaller than the SARIMA model's 99.44, 79.15, and 7.86% respectively. The RMSE, MAE, and MAPE of the forecasting effect were 100.86, 75.94, 9.57%, which were all smaller than 125.44, 97.33, 10.89% of the SARIMA model.  Conclusions  The SARIMA-GRNN combined model has a better fitting effect and higher forecasting effect than the traditional SARIMA model to forecast the monthly incidence of typhoid fever and paratyphoid fever in China. It can be used to predict the monthly incidence of typhoid fever and paratyphoid fever.