基于四种时间序列模型对西安市食源性疾病发病率预测的应用与比较  

Application and comparison of four time-series models in predicting the incidence of foodborne disease in Xi′an City

作　　者：高凡 张辉[1] 刘冬[1] 张锋[1] 刘萍[1] GAO Fan;ZHANG Hui;LIU Dong;ZHANG Feng;LIU Ping(Xi′an Center for Disease Control and Prevention,Xi′an,Shaanxi 710054,China)

机构地区：[1]西安市疾病预防控制中心,陕西西安710054

出　　处：《公共卫生与预防医学》2025年第3期12-16,共5页Journal of Public Health and Preventive Medicine

基　　金：西安市卫生科研项目(J201802036)。

摘　　要：目的 比较四种时间序列模型在西安市食源性疾病发病率中的预测应用,旨在为西安市食源性疾病防控提供科学参考。方法 收集2017年7月至2023年12月西安市食源性疾病月度发病率数据,以2017年7月至2023年6月发病率数据为训练集,分别建立季节性差分自回归移动平均(seasonal autoregressive integrated moving average,SARIMA)模型、Holt-Winters指数平滑模型、神经网络自回归(neural network autoregressive,NNAR)模型和长短期记忆(long short-term memory,LSTM)神经网络模型。以2023年7~12月食源性疾病发病率作为测试集,采用均方根误差(RMSE)、平均绝对误差(MAE)和平均绝对百分比误差(MAPE)比较模型拟合预测效果。结果 西安市2017年7月至2023年6月食源性疾病月发病率有明显的季节性特征,每年6~8月为发病高峰。SARIMA、Holt-Winters、NNAR模型、LSTM模型预测的RMSE、MAE和MAPE分别为2.88、2.43、23.71%,2.14、1.77、18.06%,3.26、2.77、26.92%和1.63、1.27、12.17%。预测效果最好的为LSTM模型。结论 LSTM模型预测准确度相对最高,可用于西安市食源性疾病发病率的短期预测。Objective To compare the application of four time-series models in the prediction of the foodborne disease incidence in Xi′an City,and to provide scientific reference for the prevention and control of foodborne diseases in Xi′an.Methods The foodborne disease monthly incidence data of Xi′an City from July 2017 to December 2023 were collected.The incidence data from July 2017 to June 2023 were used as training set,to build the seasonal autoregressive integrated moving average(SARIMA) model,Holt-Winters exponential smoothing model,neural network autoregressive(NNAR) model and long short-term memory(LSTM) neural network model.The data concerning the incidence rates of foodborne diseases from July to December 2023 were used as a test set.The predictive value of each model was evaluated using root mean squared error(RMSE),mean absolute error(MAE) and mean absolute percentage error(MAPE).Results The monthly incidence of foodborne diseases in Xi′an City from July 2017 to June 2023 showed a significant seasonal distribution characteristic,with the peak of onset from June to August each year.The RMSE,MAE and MAPE predicted by the four models were 2.88,2.43,23.71%;2.14,1.77,18.06%;3.26,2.77,26.92%;and 1.63,1.27,12.17%,respectively.The time-series model with the best prediction effect was the LSTM model.Conclusion The LSTM model produces better prediction accuracy,and can be used to predict short-term trend of foodborne disease incidence in Xi′an City.

关 键 词：食源性疾病 SARIMA模型 Holt-Winters模型 NNAR模型 LSTM模型 

分 类 号：R181[医药卫生—流行病学]