| Influenza is an acute respiratory infectious disease caused by influenza virus.The outbreak and spread of seasonal influenza each year seriously threaten human health and cause significant economic losses.If the influenza virus mutates greatly,it will cause more serious health hazards.Especially in densely populated modern urban areas,the production activities are mainly industrial and service industries,which are often accompanied by the phenomenon of dense population flow,which provides suitable conditions for the spread of influenza virus.Once the influenza epidemic breaks out in cities,it will not only endanger the safety of the people.especially the elderly,children and other high-risk groups.but also cause social impacts such as school suspension.Due to the rapid and continuous mutation of influenza viruses,in order to prevent influenza pandemics for a long time,researchers have constructed a large number of short-term prediction models of influenza epidemics through data-driven statistical learning methods to support accurate,timely and rapid epidemic situation predictions.It is of great significance to predict the risk of influenza spread in real time and conduct air defense deployment in advance.However,most of the existing data-driven short-term trend prediction studies play a role at the global,national,provincial/state or city level,which is difficult to support the refined prediction of the development trend of influenza in each region within the city,leading to the difficulty in providing differentiated intervention measures within the city.In addition,most of the existing data driven outbreak tend to rely on short-term trend prediction method of influenza-like illness statistical data to predict the time series characteristics,without considering the flu activity spatial correlation between adjacent areas,such methods as the elaborating the lack of spatial information is hard to predict the short-term trend of the development of the disease within the city.Although some prediction methods make comprehensive use of spatial and temporal characteristics of specific locations,spatial feature extraction methods ignore the impact of population movement and interaction within cities on the spread and spread of infectious diseases,which needs to be improved.In view of these problems existing in existing methods,this study proposes a spatiotemporal prediction model of intra-city influenza epidemics with a deep neural network coupled with spatial interaction.The spatiotemporal prediction model uses a home-based flow to build a relational graph network,which can more effectively capture the cross-regional movement and interaction patterns of people,and thus achieve an effective simulation of the spread and spread of influenza epidemics in cities.On this basis,based on deep learning methods,the spatial and temporal dependencies in influenza epidemic surveillance data were modeled.In this study,experiments were carried out using the Shenzhen Influenza Epidemic Surveillance Dataset.The experimental results show that,compared with the baseline model that does not consider the movement and interaction patterns of people in the city,the short-term prediction method of influenza epidemics proposed in this study has achieved better prediction performance,and can realize the prediction of influenza development trend at the street level in the city.At the same time,the ablation experiments prove that the home-based traffic-based relationship graph construction method can improve the prediction performance of the spatiotemporal prediction model more effectively than other graph network construction methods.To sum up,this study proposes a modeling method for intra-city spatial interaction suitable for influenza.Based on the deep learning model,a higher spatial resolution prediction of the development trend of intra-city influenza can be achieved.It helps the government and public health departments to predict the risk of influenza spread in real time and make prevention and control deployment in advance. |
