Research On Infectious Disease Prediction Model Based On COVID-19 Transmission Characteristic

Since the outbreak of COVID-19,the disease has rapidly spread globally within a short period of time.According to relevant studies,the novel coronavirus has unique transmission characteristics.The virus not only has a latent period of several days,but also is infectious during this period.In addition,there is a type of hidden latent carrier in the process of virus transmission.These individuals may not show symptoms after the latent period,but they can still transmit the virus.To reveal the intrinsic transmission mechanism of COVID-19 and predict its transmission trend,this paper mainly focuses on the following work:Firstly,a SEAIHR dynamic model was proposed by analyzing the transmission characteristics of COVID-19.The model takes into account the self-healing process of hidden latent individuals and the impact of early isolation of latent individuals on epidemic transmission,introducing "hidden healing state" and "hospital isolation state".Meanwhile,the influence of prevention and control measures is considered by introducing "symptomatic state".Based on the real epidemic data in Wuhan,the model considers the changes in parameters at different stages with the development of the epidemic and prevention and control measures.The basic reproductive number is calculated,and different model comparison experiments are conducted.The experimental results show that the SEAIHR model has significantly improved the fitting accuracy compared with the classical SIR and SEIR models,reducing the fitting error by 34.4%-72.8% in the early and midstages of the epidemic and better revealing the transmission mechanism of COVID-19.In addition,taking Xinjiang as an example,prediction experiments of the model were conducted.The model parameters were determined by fitting the data,and the inflection point and peak confirmed cases were predicted.In addition,a neural network model that combines recurrent neural networks and attention mechanism is proposed.In order to more accurately predict the development trend of the epidemic,this paper transforms the prediction of the trend of infectious diseases into a time-series prediction problem considering long-term dependence.By studying the phased development and changes of the COVID-19 epidemic,as well as the implementation and adjustment of corresponding prevention and control measures,the key influencing factors of the epidemic are analyzed and collected.Shanghai epidemic data is used for experiments,and different neural network structures are used for model training and prediction.The results show that the bidirectional long short-term memory network combined with self-attention mechanism can better excavate the temporal relationship of epidemic data and has significant performance improvement.What’s more,a SEAIHRBi LSTM fusion model is proposed by combining a mixed neural network with the SEAIHR model for linear regression.The optimal values of various parameters in the model were determined through multiple experiments with the mixed neural network.Based on the2022 Shandong data,the model was trained and tested,and prediction methods such as single warehouse model,SEIR-LSTM model,and hybrid neural network were compared.The experimental results showed that compared with other methods,the proposed fusion model has good analytical ability for complex epidemic development situations and more accurate and reliable prediction results.