Spatio-temporal Analysis Of COVID-19 Epidemic In China And Neural Network-based Epidemic Prediction Models

Since the outbreak of the novel coronavirus in Wuhan in late 2019,the virus has rapidly spread around the world and led to a "global pandemic",causing great harm to the health and economy of people around the world.It has also surpassed the SARS epidemic in 2002 and become the most serious public health event in China since the founding of New China.Since the introduction of the mutant strain "Omicron" into China at the end of 2021,large-scale outbreaks of COVID-19 occurred in Jilin Province and Shanghai in the first half of 2022,posing new challenges to China’s epidemic prevention policies and public health planning.Therefore,this paper conducted a detailed spatio-temporal analysis of the transmission process of COVID-19 in China during January-June 2022,and established a prediction model for daily newly confirmed cases.This study collected time series data sets of COVID-19 from 31 provincial-level administrative regions(excluding Hong Kong,Macao and Taiwan)and 333 prefecturelevel administrative regions in China for analysis,including data on confirmed COVID-19 cases,meteorological data,air quality data,policy index data and Baidu migration data.Firstly,global spatial autocorrelation and local spatial autocorrelation methods were used to explore the spatial and temporal clustering and other spatial epidemiological characteristics of COVID-19.Secondly,Spearman correlation analysis and multivariate analysis were used to explore the relationship between natural and social factors and the daily newly confirmed cases,and statistically significant influencing factors were screened out.Finally,the WNN-NARX hybrid model was constructed and compared with the BP neural network model,LSTM model and Informer model,and the screened variables were incorporated into the model to model and forecast the COVID-19 epidemic in China,Jilin Province and Shanghai.The results of this study are as follows:(1)March-May 2022 is the peak period for outbreaks in China,especially in Jilin Province from March to April and Shanghai from April to May,both of which have experienced large outbreaks.Since early June,the epidemic has been moderating across the country.(2)In terms of the global Moran’s I index,there was a significant spatial autocorrelation(p < 0.001,Z score > 7.15,Moran’s I> 0)for newly confirmed COVID-19 cases in the first half of 2022.According to the clustering and outlier analysis results,Xi ’an,Anyang and Tianjin in January-February 2022,Nantong and Suzhou in AprilJune 2022 all belong to the "low-high" clustering region,while Zhengzhou and Beijing in January-February 2022,Changchun,Jilin and its surrounding cities in March 2022 belong to the "high-high" clustering region.According to the results of Getis-Ord G *,Xi ’an City,Tianjin City,Anyang City and their surrounding cities in January 2022,Jilin City and Changchun City and their surrounding cities in March,and Shanghai City and its surrounding cities in April have successively become extremely important hotspots(99% Confidence).(3)Analysis of risk factors for COVID-19 transmission through social and natural factors showed that in the outbreak areas,such as Jilin Province and Shanghai,humidity,migration scale index and the number of daily confirmed cases were strongly negatively correlated,and ozone concentration was positively correlated;There was a strong negative correlation between confirmed cases and migration intensity index,migration intensity index and leisure dining and leisure travel intensity index.Among the social factors,the policy index stringency index,stringency legacy index,government response index and control health index showed strong positive correlation with confirmed COVID-19 cases.After the collinearity factor was removed,ozone concentration,strict legacy index,controlled health index and dining leisure travel index were included in the subsequent epidemic prediction model.(4)The WNN-NARX model,BP neural network model,LSTM model and Informer model were used to model and predict confirmed cases in China,Jilin Province and Shanghai respectively.Compared with the four models,the WNN-NARX model improved the prediction accuracy on the basis of the NARX model.Compared with BP neural network and Informer model,it can predict the trend of epidemic and the number of confirmed cases better.Compared with the LSTM model,although the prediction accuracy is lower than that of the LSTM model,the fitting effect of the training set is better.This study analyzed the spatio-temporal clustering and spatial dependence of COVID-19 in China since 2022 from multiple perspectives and used a variety of methods to simulate the spread of COVID-19 in China,providing a scientific basis for the assessment of COVID-19 hazards and the dynamic planning of public health resources in the new stage of epidemic prevention.