ARIMA模型和Holt-Winters指数平滑法在贵州省肺结核发病预测中的应用

Application of ARIMA and Holt-Winters exponential smoothingin the prediction of tuberculosis in Guizhou Province

目的探讨比较自回归差分移动平均(autoregressive integrated moving average, ARIMA)模型及Holt-Winters指数平滑法在肺结核发病预测中的应用,为贵州省结核病防控工作提供科学依据。方法以2015—2021年贵州省登记报告的肺结核发病数据建立ARIMA模型和Holt-Winters指数平滑模型,分别用两种模型预测2022年1—10月肺结核发病数,并与实际登记报告肺结核发病数作比较,对两种模型预测效果进行评价。结果 2015—2021年贵州省累计报告肺结核患者28.08万例,总体上肺结核发病呈下降趋势。构建ARIMA最佳模型为ARIMA (1, 1, 0)(0, 1, 0)12,模型预测结果均方根误差(root mean square error,RMSE)为462.46,平均绝对误差(mean absolute error,MAE)为424.50,平均绝对百分比误差(mean absolute percentage error,MAPE)为21.21%;Holt-Winters指数平滑法最佳模型为乘法模型,模型预测结果 RMSE为387.01,MAE为344.20,MAPE为17.44%。结论 Holt-Winters指数平滑模型预测效果优于ARIMA模型,更适合对贵州省肺结核发病情况进行短期预测。

Objective To explore and compare the application of ARIMA model and Holt-Winters exponential smoothing method in the prediction of tuberculosis incidence,in order to provide scientific evidence for the prevention and control of tuberculosis in Guizhou province.Methods An ARIMA model and a Holt-Winters exponential smoothing model were established based on the incidence data of pulmonary tuberculosis registered and reported in Guizhou province from 2015 to 2021 to predict the number of incident cases of pulmonary tuberculosis from January to October 2022.The prediction of the two models were compared with the actual number of incident pulmonary tuberculosis registered and reported in the target period,so as to evaluate the performance of the two models.Results From 2015 to 2021,a total of 280,822 patients with tuberculosis were reported in Guizhou Province.The incidence of tuberculosis showed a downward trend during that period.The optimal ARIMA model was ARIMA(1,1,0)(0,1,0)12,with root mean square error(RMSE)of 462.46,mean absolute error(MAE)of 424.50 and mean absolute percentage error(MAPE)of 21.21%.The optimal model of Holt-Winters exponential smoothing was multiplicative model,with RMSE of 387.01,MAE of 344.20 and MAPE of 17.44%.Conclusions The Holt-Winters exponential smoothing method may performe better than ARIMA model in predicting the incidence of pulmonary tuberculosis in Guizhou Province.