ARIMA-MLP与ARIMA-RBF模型在流行性腮腺炎发病预测中的应用  

Application of ARIMA-MLP and ARIMA-RBF model on the prediction of mumps epidemic

作　　者：曹磊[1] 张义[1] 刘峰[1] 邓勇[1] 陈飒[1] 张志成[1] 周体操[1] 

机构地区：[1]陕西省疾病预防控制中心,西安710054

出　　处：《公共卫生与预防医学》2016年第2期26-30,共5页Journal of Public Health and Preventive Medicine

摘　　要：目的探索流行性腮腺炎月发病数的最佳预测模型,为流腮发病的预测和预警提供理论依据。方法使用SPSS18.0软件,分别采用单纯自回归移动平均模型(ARIMA模型)、自回归移动平均-多层感知器神经网络模型(ARIMA-MLP组合模型)及自回归移动平均-径向基函数神经网络模型(ARIMA-RBF组合模型)对陕西省2009—2014年流腮月发病数进行拟合,找出最佳预测模型。结果单纯ARIMA模型拟合优度R^2值为0.831,平均绝对误差(MAE)值为267.49;ARIMA-MLP组合模型的R^2值为0.881,MAE值为208.01;ARIMA-RBF组合模型的R^2值为0.898,MAE值为205.82。结论 ARIMARBF组合模型对陕西省流腮月发病数预测效果最佳,可以为流腮发病的预测、预警提供理论依据。Objective To explore the best prediction model for monthly incidence of mumps,and provide theoretical basis for the prediction and early warning of mumps prevalence. Methods SPSS18. 0 software was applied,Autoregressive Integrated Moving Average Model（ ARIMA model）,Autoregressive Integrated Moving Average-Multilayer Perceptron Model（ ARIMA-MLP combined model） and Autoregressive Integrated Moving Average-Radial Basis Function Model（ ARIMA-RBF combined model） were used to fit the incidence of mumps in Shaanxi province from 2009 to 2014,and find the optimal prediction model. Results The goodness of fit of ARIMA model,R^2 was 0. 831,mean absolute error（ MAE） was 267. 49; the R^2 of ARIMA-MLP combined model was 0. 8811,MAE was 208. 01; the R^2 of ARIMA-RBF combined model was 0. 898,MAE was 205. 82. Conclusion The simulation effect of ARIMA-RBF combined model for the mumps incidence prediction of Shaanxi Province was the best,it could provide theoretical basis for the prediction and early warning of mumps prevalence.

关 键 词：流行性腮腺炎 自回归移动平均模型 多层感知器神经网络模型 径向基函数神经网络模型 

分 类 号：R512.1[医药卫生—内科学]