Research And Implementation Of COVID-19 Epidemic Forecasting Method Based On Spatio-Temporal Characteristics Mining

The widespread global spread of COVID-19 has had a significant impact on the economic,social and people ’s daily lives worldwide.Effective prediction of epidemic trends is of great significance for guiding epidemic prevention and control.The real-time collection and publication of epidemic data prompted a large number of researchers to establish an epidemic forecasting model based on data-driven methods.At present,most studies use epidemic time series data to model the overall development trend of the epidemic,and rarely consider the spatio-temporal characteristics of the spread of COVID-19 in multiple regions.However,the epidemic has a high spatial correlation among regions,and has different propagation characteristics at different time stages,which leads to complex spatio-temporal dependence in the epidemic data,and limits the expression ability and prediction effect of the current epidemic forecasting model.In view of a series of challenges brought by the spatial correlation of the spread of the epidemic and the difference of growth patterns in different time ranges,this thesis carried out the research and implementation of the COVID-19 epidemic forecasting method based on spatio-temporal characteristics mining.First of all,in view of the spatial correlation of the outbreak in multiple regions,a regional epidemic forecasting method based on spatial correlation perception is proposed.This method not only uses a variety of correlation information between regions to model the correlation between epidemic spread in multiple regions but also captures the potential correlation between epidemic spread in multiple regions when predicting the epidemic trend in multiple regions synchronously,and introduces the influence of the spatial correlation of epidemic spread on the epidemic trend.Secondly,in view of the complex temporal dependence of the epidemic at different time stages,the difference of the epidemic temporal pattern in different time ranges was verified by clustering analysis of the epidemic time series data,and the epidemic forecasting method based on temporal pattern recognition was proposed.This method carried out temporal pattern recognition on the longterm trend and the short-term local trend affecting the development of the epidemic in different time ranges respectively.In the process of epidemic trend forecasting,the influence of the epidemic temporal pattern in different time ranges on the overall trend of the epidemic was integrated.In this study,a number of COVID-19 epidemic datasets in the real world are used to verify and analyze the proposed epidemic forecasting methods.The experimental results under multiple evaluation indexes show that the effect of the proposed epidemic forecasting methods is better than that of the commonly used benchmark methods.The regional epidemic forecasting model based on spatial correlation perception can capture the strength of the spatial association of epidemic transmission in multiple regions,and provide reference for the joint prevention and control of epidemics in multiple regions.The epidemic forecasting model based on temporal pattern recognition can effectively integrate the influence of various epidemic temporal patterns on the epidemic trend and improve the prediction ability of the model.To verify the feasibility of the proposed epidemic forecasting methods,the thesis designs and implements a COVID-19 forecasting system.The system can automatically collect epidemic data,visually analyze the real-time situation of the epidemic,simulate the spread process of the epidemic,predict the future development trend of the epidemic,and provide epidemic prevention and control guidance for users’ daily life,which has high application value.