BPNN神经网络模型和SARIMA模型在荆州市乙类传染病发病数中的预测效果比较  

A Comparison of the Prediction Effects of Back Propagation Neural Network Model and Seasonal Autoregressive Integrated Moving Average Model in the Caseload of Class B Notifiable Diseases in Jingzhou City

作　　者：刘天[1] 姚梦雷[1] 黄继贵[1] 黄淑琼[2] 陈红缨[2] 杨雯雯[2] 蔡晶[2] 吴然 LIU Tian;YAO Menglei;CHEN Hongying(Jingzhou Municipal Center for Disease Control and Prevention,Jingzhou,Hubei,434000,China;不详)

机构地区：[1]荆州市疾病预防控制中心,湖北荆州434000 [2]湖北省疾病预防控制中心,湖北武汉430079

出　　处：《中国社会医学杂志》2021年第1期109-113,共5页Chinese Journal of Social Medicine

基　　金：湖北省卫生计生委创新团队项目(WJ2016JT-002)。

摘　　要：目的评价BPNN神经网络模型和季节性差分自回归滑动平均模型(seasonal autoregressive integrated moving average,SARIMA)在乙类传染病发病数中的预测效果。方法利用荆州市2005年1月—2017年12月的乙类传染病逐月发病数作为拟合数据,建立BPNN神经网络模型和SARIMA模型,预测2018年1—5月逐月发病数并与实际值比较,采用平均绝对百分比误差(MAPE)、R^(2)、均方误差(RMSE)和平均绝对误差(MAE)评价模型的拟合及预测效果。结果SARIMA模型建立的最优模型为SARIMA[0,1,(12)](1,1,1)12,且残差为白噪声序列。BPNN神经网络模型和SARIMA模型拟合的MAPE、R2、RMSE和MAE依次分别为3.92%,0.92,82.29,61.93;7.16%,0.49,149.93,118.10。BPNN神经网络模型和SARIMA模型预测的MAPE、R2、RMSE和MAE依次分别为11.84%,0.23,180.33,94.76;21.96%,-0.91,633.94,251.19。结论BPNN神经网络模型对荆州市乙类传染病发病数拟合和预测效果均优于SARIMA模型。Objective Evaluating the effect of the Back Propagation Neural Network(BPNN)model and Seasonal Autoregressive Integrated Moving Average(SARIMA)model in the prediction of class B notifiable diseases in Jingzhou City.Methods Data of class B notifiable diseases monthly caseload from January 2005 to December 2017 in Jingzhou City were used to establish BPNN model and SARIMA model,and predicted the monthly cases between January and April 2018 and compared with the actual value.Mean absolute percentage error(MAPE),R^(2),root mean square error(RMSE)and mean absolute error(MAE)were used to evaluate the fit and prediction effects.Results The ARIMA[0,1,(12)](1,1,1)12 model was the most appropriate one with the residual test showing a white noise sequence.In the fitting phase,the MAPE,R2,RMSE and MAE fitted by BPNN model and SARIMA model were 3.92%,0.92,82.29,61.93 and 7.16%,0.49,149.93,118.10.The MAPE,R2,RMSE and MAE predicted by BPNN model and SARIMA model were 11.84%,0.23,180.33,94.76 and 21.96%,-0.91,633.94,251.19.Conclusion The BPNN model showed better class B notifiable diseases fitting and forecasting in Jingzhou City than the SARIMA model.

关 键 词：BPNN神经网络模型 SARIMA模型 乙类传染病 预测 

分 类 号：R183[医药卫生—流行病学]