BP人工神经网络模型在上海市感染性腹泻日发病例数预测中的应用  

Preliminary application of Back-Propagation artificial neural network model on the prediction of infectious diarrhea incidence in Shanghai

作　　者：黎健[1] 顾君忠[2] 毛盛华[1] 肖文佳[1] 金汇明[1] 郑雅旭[1] 王永明[2] 胡家瑜[1] 

机构地区：[1]上海市疾病预防控制中心,200336 [2]华东师范大学计算机应用研究所

出　　处：《中华流行病学杂志》2013年第12期1198-1202,共5页Chinese Journal of Epidemiology

基　　金：上海市公共卫生重点学科计划（12GWZX0101）

摘　　要：目的建立基于气象因素的上海市感染性腹泻逐日发病例数BP人_T神经网络预测模型。方法收集l：海市2005--2008年感染性腹泻逐日发病例数与同期气象资料包括气温、相对湿度、降雨量、气压、日照时数、风速，通过Spearman相关分析选出与感染性腹泻相关的气象因素，用主成分分析（PCA）去除气象因素间的共线性影响。利用MatLabR2012b软件的神经网络工具箱建立感染性腹泻日发病例数的BP神经网络预测模型，并对拟合效果、外推预测能力和等级预报效果进行评价。结果Spearman相关性分析显示，日感染性腹泻与前两天的日最高气温、最低气温、平均气温、最低相对湿度、平均相对湿度呈正相关（P〈O．01），与前两天的日平均气压呈负相关（P〈0．01）。输入PCA提取的4个气象主成分构建BP神经网络预测模型，训练和预测样本平均绝对误差、均方根误差、相关系数、决定系数分别为4．7811、6．8921、O．7918、0．8418和5．8163、7．8062、0．7202、0．8180。模型预测值对2008年实际发病数的年平均误差率为5．30％，对感染性腹泻的等级预报正确率为95．63％。结论温度和气压对感染性腹泻日发病例数影响较大。BP神经网络模型的拟合及预测误差较小，预报正确率较高，预报效果理想。Objective To establish BP artificial neural network predicting model regarding the daily cases of infectious diarrhea in Shanghai. Methods Data regarding both the incidence of infectious diarrhea from 2005 to 2008 in Shanghai and meteorological factors including temperature, relative humidity, rainfall, atmospheric pressure, duration of sunshine and wind speed within the same periods were collected and analyzed with the MatLab R2012b software. Meteorological factors that were correlated with infectious diarrhea were screened by Spearman correlation analysis. Principal component analysis（PCA） was used to remove the multi-colinearities between meteorological factors. Back-Propagation （BP） neural network was employed to establish related prediction models regarding the daily infectious diarrhea incidence, using artificial neural networks toolbox. The established models were evaluated through the fitting, predicting and forecasting processes. Results Data from Spearman correlation analysis indicated that the incidence of infectious diarrhea had a highly positive correlation with factors as daily maximum temperature, minimum temperature, average temperature, minimum relative humidity and average relative humidity in the previous two days （P〈0.01）, and a relatively high negative correlation with the daily average air pressure in the previous two days （P〈 0.01 ）. Factors as mean absolute error, root mean square error, correlation coefficient （r） , and the

关 键 词：感染性腹泻 气象因素 BP人工神经网络 

分 类 号：R181.3[医药卫生—流行病学] TP183[医药卫生—公共卫生与预防医学]