应用Elman神经网络建立流感样病例预测模型
|
摘要
目的应用Elman神经网络构建流行性感冒(流感)样病例(ILI)预测模型,为浙江省流感疫情早期预警提供依据。方法收集2013—2014年浙江省11家流感监测哨点医院的ILI报告、ILI病毒核酸检测结果、气象和空气污染物等资料,通过时滞相关性分析筛选纳入模型的变量,采用2013年第14周—2014年第44周的数据建立Elman神经网络预测模型,采用2014年第45—52周的数据检验模型的预测效能。结果浙江省2013—2014年每周均有ILI报告,共报告506 391例次,周报告ILI%为(3.07±0.73)%。筛选出提前13周的周平均气压、提前11周的周平均水汽压、提前9周的周平均气温、提前5周的周平均SO2浓度、提前5周的周平均NO2浓度、提前5周的周平均CO浓度、提前5周的周平均PM_(2.5)浓度、提前5周的周平均PM_(10)浓度、提前5周的周平均AQI和提前1周的病原阳性率10个因素纳入模型。当网络结构为10-15-1-1时,构建的Elman模型为最优预测模型,预测结果的平均误差绝对率为10.58%,非线性相关系数为0.876 7。结论利用气象、空气污染指标和流感病原性监测资料建立的Elman神经网络ILI预测模型预测效果较好,适用于浙江省流感疫情短期预测。
Objective To build a model for influenza-like illness(ILI)prediction based on Elman neural network and to provide evidence for early warning of influenza epidemic in Zhejiang Province. Methods The data of ILI from 11 sentinel hospitals, influenza pathogen detection, meteorological factors and air pollutants in Zhejiang Province from 2013 to 2014 were collected. Time-delay correlation analysis was conducted to select variables for modeling. Based on Elman neural network,data from the 14 th week of 2013 to the 44 th week of 2014 were used as a training set to establish the model and the data from 45 th week to 52 nd weeks of 2014 were used as a test set for the model performance. Results There were ILI reported every week during 2013 and 2014, with a total of 506 391. The percentage of ILI cases per week was(3.07 ±0.73)%. Ten variables selected by time-delay correlation analysis were the weekly average values of atmospheric pressure(13weeks in advance),vapor pressure(11 weeks in advance),temperature(9 weeks in advance),SO2(5 weeks in advance),NO2(5 weeks in advance),CO(5 weeks in advance),PM_(2.5)(5 weeks in advance),PM_(10)(5 weeks in advance),air quality index(5weeks in advance)and positive rate of pathogen(1 weeks in advance). Elman neural network(10-15-1-1)was selected as the optimal model, and the prediction performed well, with 10.58% as the mean error rate and 0.876 7 as the nonlinear correlation coefficient. Conclusion This study demonstrated that Elman neural network including variables of meteorological factors,air pollutants and the positive rate of pathogen performed well on the short-term prediction of ILI incidence.
Objective To build a model for influenza-like illness(ILI)prediction based on Elman neural network and to provide evidence for early warning of influenza epidemic in Zhejiang Province. Methods The data of ILI from 11 sentinel hospitals, influenza pathogen detection, meteorological factors and air pollutants in Zhejiang Province from 2013 to 2014 were collected. Time-delay correlation analysis was conducted to select variables for modeling. Based on Elman neural network,data from the 14 th week of 2013 to the 44 th week of 2014 were used as a training set to establish the model and the data from 45 th week to 52 nd weeks of 2014 were used as a test set for the model performance. Results There were ILI reported every week during 2013 and 2014, with a total of 506 391. The percentage of ILI cases per week was(3.07 ±0.73)%. Ten variables selected by time-delay correlation analysis were the weekly average values of atmospheric pressure(13weeks in advance),vapor pressure(11 weeks in advance),temperature(9 weeks in advance),SO2(5 weeks in advance),NO2(5 weeks in advance),CO(5 weeks in advance),PM_(2.5)(5 weeks in advance),PM_(10)(5 weeks in advance),air quality index(5weeks in advance)and positive rate of pathogen(1 weeks in advance). Elman neural network(10-15-1-1)was selected as the optimal model, and the prediction performed well, with 10.58% as the mean error rate and 0.876 7 as the nonlinear correlation coefficient. Conclusion This study demonstrated that Elman neural network including variables of meteorological factors,air pollutants and the positive rate of pathogen performed well on the short-term prediction of ILI incidence.
引文
[1] PAUL S, MGBERE O, ARAFAT R, et al. Modeling and forecasting influenza-like illness(ILI)in Houston,Texas using three surveillance data capture mechanisms[J]. Online journal of public health informatics,2017,9(2):e187.
[2]谢亮,富小飞,亓云鹏,等.嘉兴市流行性感冒监测分析[J].预防医学,2017,29(8):806-808.
[3]陈健,张磊,陆帅,等.采用广义可加模型分析预测上海市流感样病例发病情况[J].中华流行病学杂志,2013,34(4):404-408.
[4]杨召,叶中辉,赵磊,等. ARIMA-BPNN组合预测模型在流感发病率预测中的应用[J].中国卫生统计,2014,31(1):16-18.
[5]王斌,冯慧芬,黄平,等.人工神经网络模型在预测重症手足口病中的应用研究[J].现代预防医学, 2018, 45(11):1921-1924,1947.
[6] NIU Z,JIA Y,ZHANG L,et al. Prediction for short-term traffic flow based on Elman neural network optimized by CPSO[J].Metallurgical&Mining Industry,2015(9):991-995.
[7] CHU C, CHANG C, CHANG T, et al. Elman neural network identify elders fall signal base on second-order train method[C]//2017 6th International Symposium on Next Generation Electronics(ISNE),2017:1-4.
[8]黄薇,刘建平,张锦周,等.细菌性食源性疾病发病预测中Elman神经网络模型的应用[J].公共卫生与预防医学,2013,24(6):10-13.
[9]孙锦峰,耿云亮,郭奕瑞,等. Elman神经网络与ARIMA模型对流感发病率预测效果的比较[J].郑州大学学报(医学版),2013,48(5):584-587.
[10]中华人民共和国卫生部.关于印发《全国流感监测方案(2010年版)》的通知[Z].(2010-09-10)[2018-08-24]. http://www.nhfpc.gov.cn/jkj/s3577/201009/3fa356d0f4834d408fde6c12891 a6482.shtml.
[11]何树红,杨博,戴明爽,等.基于动态递归神经网络的石油价格预测[J].云南民族大学学报(自然科学版),2013,22(1):33-35.
[12]黎漓.湖南省流行性腮腺炎的流行特征及Elman神经网络预测[D].衡阳:南华大学,2015.
[13]王芳,涂春丽,勾永尧.基于Elman神经网络的气温预测研究[J].安徽农业科学,2011,39(33):20859-20860.
[14] KOELLE K,RASMUSSEN D A. Influenza:prediction is worth a shot[J]. Nature,2014,507(7490):47-48.
[15]邓斐,资海荣,余慧燕,等.江苏省2010—2014年流感流行特征分析[J].中华疾病控制杂志,2016,20(9):901-904.
[16]李明,冯录召,曹玉,等.中国2005—2013年流感暴发疫情的流行病学特征分析[J].中华流行病学杂志, 2015, 36(7):705-708.
[17]王伟. 2009—2014年山东省流感样病例流行特征及其与气象因素关系研究[D].济南:山东大学,2016.
[18]杨旭辉,王婧,孙昼,等.气象因子与流感样病例的相关性研究[J].浙江预防医学,2014,26(7):653-658.
[19] SOEBIYANTO R P, CLARA W A, JORGE J, et al.Associations between seasonal influenza and meteorological parameters in Costa Rica,Honduras and Nicaragua[J]. Geospatial Health,2015,10(2):372.
[20]许国章,王春丽,李永东,等. 2008—2015年宁波市流感样病例预测模型分析[J].国际流行病学传染病学杂志,2016,43(1):30-34.
[2]谢亮,富小飞,亓云鹏,等.嘉兴市流行性感冒监测分析[J].预防医学,2017,29(8):806-808.
[3]陈健,张磊,陆帅,等.采用广义可加模型分析预测上海市流感样病例发病情况[J].中华流行病学杂志,2013,34(4):404-408.
[4]杨召,叶中辉,赵磊,等. ARIMA-BPNN组合预测模型在流感发病率预测中的应用[J].中国卫生统计,2014,31(1):16-18.
[5]王斌,冯慧芬,黄平,等.人工神经网络模型在预测重症手足口病中的应用研究[J].现代预防医学, 2018, 45(11):1921-1924,1947.
[6] NIU Z,JIA Y,ZHANG L,et al. Prediction for short-term traffic flow based on Elman neural network optimized by CPSO[J].Metallurgical&Mining Industry,2015(9):991-995.
[7] CHU C, CHANG C, CHANG T, et al. Elman neural network identify elders fall signal base on second-order train method[C]//2017 6th International Symposium on Next Generation Electronics(ISNE),2017:1-4.
[8]黄薇,刘建平,张锦周,等.细菌性食源性疾病发病预测中Elman神经网络模型的应用[J].公共卫生与预防医学,2013,24(6):10-13.
[9]孙锦峰,耿云亮,郭奕瑞,等. Elman神经网络与ARIMA模型对流感发病率预测效果的比较[J].郑州大学学报(医学版),2013,48(5):584-587.
[10]中华人民共和国卫生部.关于印发《全国流感监测方案(2010年版)》的通知[Z].(2010-09-10)[2018-08-24]. http://www.nhfpc.gov.cn/jkj/s3577/201009/3fa356d0f4834d408fde6c12891 a6482.shtml.
[11]何树红,杨博,戴明爽,等.基于动态递归神经网络的石油价格预测[J].云南民族大学学报(自然科学版),2013,22(1):33-35.
[12]黎漓.湖南省流行性腮腺炎的流行特征及Elman神经网络预测[D].衡阳:南华大学,2015.
[13]王芳,涂春丽,勾永尧.基于Elman神经网络的气温预测研究[J].安徽农业科学,2011,39(33):20859-20860.
[14] KOELLE K,RASMUSSEN D A. Influenza:prediction is worth a shot[J]. Nature,2014,507(7490):47-48.
[15]邓斐,资海荣,余慧燕,等.江苏省2010—2014年流感流行特征分析[J].中华疾病控制杂志,2016,20(9):901-904.
[16]李明,冯录召,曹玉,等.中国2005—2013年流感暴发疫情的流行病学特征分析[J].中华流行病学杂志, 2015, 36(7):705-708.
[17]王伟. 2009—2014年山东省流感样病例流行特征及其与气象因素关系研究[D].济南:山东大学,2016.
[18]杨旭辉,王婧,孙昼,等.气象因子与流感样病例的相关性研究[J].浙江预防医学,2014,26(7):653-658.
[19] SOEBIYANTO R P, CLARA W A, JORGE J, et al.Associations between seasonal influenza and meteorological parameters in Costa Rica,Honduras and Nicaragua[J]. Geospatial Health,2015,10(2):372.
[20]许国章,王春丽,李永东,等. 2008—2015年宁波市流感样病例预测模型分析[J].国际流行病学传染病学杂志,2016,43(1):30-34.