摘要
目的：探讨适合全国乙肝发病率的预测模型，为乙肝预测预警系统提供参考。方法应用2004－2012年全国乙肝月发病率数据，分别建立 ARIMA 模型和 BP 神经网络模型，利用建立的模型预测2013年1－12月乙肝发病率，采用实际发病率验证与比较两种模型的预测效果，评价指标为平均绝对误差（MAE）、平均绝对误差率（MER）和非线性相关系数（RNL）。结果全国2004－2013年乙肝月发病率在2.79／10万～9.44／10万间波动，序列具有明显的长期趋势。建立的乘积 ARIMA（0，1，1）（0，1，1）12模型预测的 MAE、MER、RNL 分别为0.445、0.065、0.909，BP 神经网络模型分别为0.635、0.093、0.872。ARIMA 模型预测的平均绝对误差和平均绝对误差率要低于 BP 神经网络模型（△MAE＝0.190，△MER＝0.028），非线性相关系数要高于BP 神经网络模型（△RNL＝0.037）。结论 ARIMA 模型和BP 神经网络模型均适用于我国乙肝发病率的预测，且前者的预测效能和非线性拟合能力略优于后者。
Objective To explore suitable prediction models for hepatitis B incidence in China;to provide reference for fore-casting warning system of hepatitis B.Methods ARIMA model and Back-Propagation （BP）neural network model were estab-lished based on monthly incidence of hepatitis B from 2004 to 2012.Predication performance of both models were verified by monthly incidence of hepatitis B in 2013.Mean absolute error（MAE）,mean error rate（MER）and nonlinear correlation coeffi-cient（RNL）were used to compare prediction effects of above two models.Results The monthly incidence of hepatitis B from 2004 to 2013 were in the range of 2.79/105 -9.44/105 ,demonstrating obvious long-term trends.The MAE,MER,RNL be-tween actual values and predicted values of the monthly incidence of hepatitis B in 2013 using the fitting ARIMA（0,1,1）（0,1, 1）12 model and BP neural network model were 0.445,0.065,0.909 and 0.635,0.093,0.872,respectively.MAE and MER of ARIMA model were lower than those of BP neural network model（△MAE= 0.190,△MER= 0.028）,its RNL was high-er than that of BP neural network model（△RNL=0.037）.Conclusion Both ARIMA model and BP neural network model performed well in predicting hepatitis B incidence in China.The prediction and nonlinear fitting ability of ARIMA model was slightly better than those of BP neural network model.
出处
《江苏预防医学》
CAS
2015年第3期23-26,共4页
Jiangsu Journal of Preventive Medicine
