Prediction of COVID-19 spread by sliding mSEIR observer

Chen, Duxin; Yang, Yifan; Zhang, Yifan; Yu, Wenwu

The outbreak of COVID-19 has brought unprecedented challenges not only in China but also in the whole world. Thousands of people have lost their lives, and the social operating system has been affected seriously. Thus, it is urgent to study the determinants of the virus and the health conditions in specific populations and to reveal the strategies and measures in preventing the epidemic spread. In this study, we first adopt the long short-term memory algorithm to predict the infected population in China. However, it gives no interpretation of the dynamics of the spread process. Also the long-term prediction error is too large to be accepted. Thus, we introduce the susceptible-exposed-infected-removed (SEIR) model and further the metapopulation SEIR (mSEIR) model to capture the spread process of COVID-19. By using a sliding window algorithm, we suggest that the parameter estimation and the prediction of the SEIR populations are well performed. In addition, we conduct extensive numerical experiments to show the trend of the infected population for several provinces. The results may provide some insight into the research of epidemics and the understanding of the spread of the current COVID-19.