| Effective analysis and prediction of influenza activities in specific areas can reduce and avoid huge losses caused by influenza pandemic to a great extent.Due to the randomness and variability of influenza,and many complex factors affecting influenza activities,it is difficult to analyze and predict influenza.At present,the influenza information has to be reported,summary,analysis and released from the local sentinel hospitals to the Center for Disease Control and Prevention(CDC)level by level.This mechanism causes the delay in the release of influenza indicators such as influenza-like illness visit rate(ILI%),so as to miss the best time for prevention and control.In addition,the traditional influenza prediction models seldom fully consider the factors affecting influenza activities,resulting in a large deviation between the influenza prediction results and the actual situation.In order to solve the problems above,this paper has done the following works:(1)Grey relation analysis model for correlation analysis between ILI% and climate and environmental factors.Aiming at the complex nonlinear relationship between ILI%and climate and environmental factors,this paper construct grey relation analysis(GRA)model to quantify the impact of climate and environment on influenza.The ILI%sequence is defined as the parent sequence,and the climate and environmental factors sequence is defined as the subsequence.The mean value method is used to make the sequence dimensionless.Deng’s grey relation analysis model is established to calculate the correlation degree between ILI% and climate and environmental factors,and the importance ranking of climate and environmental factors affecting ILI% is obtained.(2)Wavelet neural network model for ILI% current week prediction.On the basis of determining the key climate and environmental factors affecting ILI% in(1),a wavelet neural network prediction model(WNN)is established to predict the current week ILI%,which can solve the problem of delayed release caused by multi sentinel information statistics.Those key factors and ILI% historical data are selected as the input,and the ILI% of the current week as the output.The wavelet neural network model is trained according to the gradient descent method,and the momentum factor is introduced in the training process to accelerate the convergence speed and prevent falling into the local minimum,so as to improve the prediction accuracy of the model.The prediction accuracy of WNN model is verified by the example of influenza-like illness visit rate in Zhejiang Province.(3)ARIMA-BRB-WNN model for ILI% short-term prediction.In order to satisfy the practical needs of influenza prevention and control,ARIMA-BRB-WNN model is proposed to predict ILI% in the short term.The predicted values of key climate and environmental factors in the next week are obtained by constructing ARIMA model.These predicted values and relevant historical data are used as the input of WNN model to obtain the predicted value of ILI% in the next week.The belief rule base(BRB)model is introduced to adaptively adjust the parameters of ARIMA model to further improve the prediction performance of the combined model.The accuracy of the combined model is verified by the example of predicting the influenza-like illness visit rate in Zhejiang Province. |
