Spatiotemporal Heterogeneity Modeling And Prediction Of COVID-19 Transmission

As a new acute respiratory infectious disease,COVID-19 poses a major threat to public health,economic life and social development all over the world.By the end of 2021,the cumulative number of confirmed cases worldwide has reached nearly 400 million,and the cumulative number of deaths has reached 5.7 million.Time,space and population are three important dimensions of epidemic transmission.Using the theories and methods of spatial statistics and deep learning,this paper explores the spatiotemporal heterogeneity and spatiotemporal dependence of epidemic transmission,and quantitatively analyzes the spatiotemporal transmission characteristics of epidemic in Hubei Province during the Spring Festival transportation in 2020 by constructing a spatiotemporal prediction model.The main work includes the following two aspects:(1)The spatiotemporal heterogeneity of the relationship between epidemic spread and population flow was analyzed by spatiotemporal geographic weighted regression model.Taking the confirmed cases of epidemic situation as the dependent variable,the urban population flow index and Wuhan population inflow index as the explanatory variables,this paper constructs a Geographically and Temporally Weighted Regression(GTWR)model.The results show that the fitting degree of GTWR model can reach more than 90%,which is better than Geographic Weighted Regression(GWR),Time Weighted Regression(TWR)and other models;Further analysis found that the relationship between epidemic spread and population mobility showed obvious phased characteristics and temporal and spatial differentiation.(2)Considering the spatiotemporal dependence,a time-space prediction model of time-series graph convolution neural network is constructed.In view of the spatiotemporal dependence in the process of covid-19 epidemic urban transmission,this paper uses the Long-Short Term Memory network(LSTM)to extract the temporal characteristics of epidemic spatiotemporal transmission,and uses the Graph Convolution Nueral network(GCN)to extract the spatial characteristics of epidemic spatiotemporal transmission,and combines the two to propose a time series graph neural network spatiotemporal prediction model.The experimental results show that the model can take into account the spatiotemporal dependence and has good spatiotemporal prediction ability.Compared with the existing similar models,it has better convergence(MAE loss value is less than33% of other models,RMSE loss value is only less than 7% of other models)and computational efficiency(training time and efficiency are more than 1.8 times faster than other models).The above experimental results show that population flow in time and space is the main driving factor of COVID-19 epidemic spread,and the spread of COVID-19 epidemic has obvious temporal and spatial characteristics.At the same time,the spatiotemporal prediction method considering the dependence of time and space can obtain better prediction effect than the traditional model.