深圳市肠道传染病与气象及媒介因素的贝叶斯正规化BP神经网络模型

目的探讨肠道传染病与相关的主要气象指标和媒介生物密度指标的关系,建立深圳市肠道传染病发病率的BP神经网络模型,并评价其拟合效果和预测效果。方法从监测系统收集2000～2005年每月的气象资料、媒介生物监测数据以及肠道传染病疫情数据,利用Stata 8.0对与肠道传染病发病率相关的气象因素及媒介生物因素进行筛选,利用Matlab 7.0软件完成贝叶斯正规化BP神经网络模型的构建、训练及模拟。结果在各种气象因素与媒介生物密度指标中,与肠道传染病发病相关程度较高的指标分别为蟑螂密度、蝇密度、平均气温和雨量。月肠道传染病发病率的回代平均误差率和R2分别为14.9%和0.87。而进行预测时,以月份为单位的肠道传染病发病率预测平均误差率为18.4%,而以年为单位的肠道传染病发病率预测的平均误差率较低,为8.4%。结论肠道传染病伤与气象因素、媒介生物因素之间关系的贝叶斯BP神经网络模型拟合效果较好,预测准确度较高,BP神经网络在传染病发病率预测研究领域具有一定的实用价值。

The Model of Bayesian-regularization Back-Propagation Neural Network about Meteorological Factors,Vector Factors and Intestinal Communicable disease in Shenzhen

Objective To investigate the relationship among meteorological factors,vector factors and intes- tinal communicable disease.We built the Back-Propagation neural network(BPNN)model and evaluated the effect of simulation and the forecast result about this model.Methods We collected the monthly data of meteorology,vec- tor and intestinal communicable disease cases from several surveillance system.We analyzed the correlation among the incidence of intestinal communicable disease,those meteorological factors and vector factors.We built the Bayesian-regularization BPNN model by Matlab 7.0.Results The valuable vector factors of incidence included the average density of roach and the average density of fly.The valuable meteorological factors was the mean of monthly air temperature and the whole monthly precipitation.The Mean Error Rate(MER)and R2 of the BPNN about monthly incidence was 14.9% and 0.87 respectively.When we forecasted the incidenee of intestinal communicable disease by the BPNN,we found that the MER of monthly incidence(18.4%)was higher than the value of annual incidence(8.4%).Conclusion The Bayesian-regularization BPNN model among meteorological factors,vector fac- tors and intestinal communicable disease offered anticipant effect in data simulation and foreeasting.BPNN have practical value Epidemiological study of communicable disease incidence estimation.