The Establishment And Simulation Of The Forecast Model Of Pertussis In Kashgar Area Of Xinjiang

Objective:To explore the feasibility of the application of one-element time series model,multivariate time series model and BP neural network model in the early warning of pertussis in Kashgar,Xinjiang,according to the actual incidence of pertussis in Kashgar,to establish a predictive model which accords with the epidemic characteristics of pertussis disease.To grasp the overall epidemic trend of pertussis in Kashgar,and to provide a feasible reference for the relevant departments to do the prevention and control work of pertussis in advance.Methods:firstly,using deterministic factor decomposition and seasonal index to analyze the trend and seasonality of pertussis incidence in Kashgar,we constructed seasonal ARIMA product seasonal model.The optimal ARIMA?p,d,q??P,D,Q?s models are determined by model identification,parameter estimation,model diagnosis and model optimization.The model was used to predict the monthly incidence data of pertussis 2011-2016 in Kashgar,and RMSE was chosen as the precision index to judge the model prediction effect.Secondly,the relationship between the incidence of pertussis and meteorological factors is analyzed by introducing meteorological factors on the basis of one-element time series model.The corresponding residual white noise sequence is obtained by establishing a one-element time series model for each variable.Based on the correlation function of residual white noise,the meteorological factors affecting the incidence of pertussis are found out,and the optimal lag time is obtained,and the selected meteorological factors are incorporated into the previously determined time series model to construct multivariate time series ARIMAX model.The optimal ARIMAX model is determined according to the minimum criterion of AIC and BIC.The research data are fitted and predicted,and the RMSE of corresponding stages are calculated.Finally,a BP neural network model was established by using the number of pertussis and meteorological data in Kashgar of Xinjiang in the last 2011-2016 years,and the influence of meteorological factors on the incidence of pertussis was studied by using meteorological factors as input and the number of pertussis as network output.This paper divides the data into training set and test set,uses the nnet package of R software to model,and estimates the parameter by Cross-validation method,and determines the optimal BP neural network model.Results:firstly,the incidence of pertussis was determined by deterministic factor decomposition and seasonal index,which was higher in July and August and least in October.According to the modeling steps of the ARIMA product seasonal model,after the smoothness test and the white noise test,the optimal model was determined by ARIMA?0,1,0??0,1,1?₁₂,corresponding to AIC=419.64,BIC=423.58.The RMSE=10.32 of the fitting stage and the prediction stage RMSE=8.77,using this model to predict the new onset number of 2017 Xinjiang Kashgar pertussis,the results showed that the incidence of pertussis in Kashgar Prefecture was lower in the 2017 year.Secondly,by drawing the CCF of the residual white noise of various meteorological factors and the incidence of pertussis,it is confirmed that the average air pressure of 4-order lag,the number of floating dust days with lag of 4-order and the number of ascending sand days with 8-order lag are positively correlated with pertussis cases,and the AIC value of the ARIMAX model with three meteorological factors is the least,AIC=394.72.The RMSE=5.03 of the fitting stage is calculated and the prediction accuracy of the model is improved obviously after the RMSE=4.28 of meteorological factors.Finally,a 3-layer BP Neural network model is established,and after repeated debugging,the RMSE value of the corresponding model is minimized when the number of hidden layer nodes is 6 and the weight decay parameter is 0.01.In the end,it is confirmed that the BP neural network model structure of this paper is 8-6-1,the model fitting stage and the prediction stage are6.14 and 4.01 respectively,the numerical value is small,which shows that the RMSE of the BP Neural network model based on the meteorological factors as the network input is reliable and reasonable.Conclusion:Using the ARIMA product seasonal model,ARIMAX model and BP neural network model can simulate the epidemic trend of pertussis in Kashgar,Xinjiang,but the multivariate time series model with meteorological factors and BP Neural network model have better predictive effect and higher precision of model prediction.The prediction results of two models can provide a theoretical basis for the relevant departments to formulate preventive measures.