Application Of Sequential Data Assimilation Methods In The Simulation And Prediction Of Epidemic Models

Data assimilation method is an optimization method that meld observations with the state transition model to improve the accuracy of prediction. Data assimilation method is divided into two main categories:continuous data assimilation method and sequential data assimilation method. Sequential data assimilation method is also known as filtering algorithm. The filtering algorithm is mainly divided into two steps:Forecast and update. First, it forecasts the state of time instant t+1 according to the state value of time instant t and state transition model; Then, it updates the predicted value by the observation of time instant t+1. So that the uncertainty in the simulation and prediction process is decreased, and we get the more realistic results.For epidemic models, the model parameters are estimated inaccurately due to the ’poor’data, or the results can not accurately reflect the realistic changes of state because some factors are ignored when we consider the model. Therefore, we consider applying the sequential data assimilation method to the model simulation and prediction to reduce the uncertainty in the simulation process, arid get more realistic results,In this paper, the research works are as follows:(1) We introduce the common methods of sequential data assimilation:Kalman filter algorithm, and Ensemble Kalman filter algorithm, and particle filter-Markov chain Monte Carlo method.(2) We apply the Kalman filter algorithm to the hepatitis B virus epidemic model simulation and prediction and get the better results.(3) We use the particle filter-Markov chain Monte Carlo method to estimate the pa-rameters and state of Richards model about Ebola epidemic, and forecast the state in a short-term, and the results are good.