城市区域功能感知的细粒度疫情风险评估模型

Fine grained epidemic risk assessment model of urban region functionality aware

针对疫情期间实施精准疫情防控的需求,达到兼顾疫情防护和社会经济发展的目的,实现了细粒度条件下的疫情风险评估深度学习模型,用于探索城市场景下街道、社区级别的疫情风险实时预测。提出一种城市区域功能感知的细粒度疫情风险评估模型,主要包括区域功能特征学习和区域关联挖掘2个模块。首先,区域功能特征学习模块将POI(point of interest)分布和疫情风险特征相融合,用于表达不同城市功能对于风险传播的影响,以引入风险扩散的先验知识。然后,区域关联挖掘模块将低层的城市网格按照功能分类映射至更高层的功能区域空间,并直接捕捉功能区域间的空间依赖,避免了低效的堆叠卷积计算。所提模型在2022年南京市新冠疫情数据集上进行了验证,相比于传统的时间序列和时空序列预测方法,所提模型相较于现有方法,在平均绝对百分比误差指标上降低了8%~23%,在均方根误差指标上降低了0.3~1.2。同时所提模型的可学习参数量大幅降低,模型计算效率远优于现有方法。

In response to the need for precise epidemic prevention and control during the pandemic,a deep learning model for fine grained epidemic risk assessment has been developed,aiming to achieve a balance between epidemic prevention and socio-economic development.This model focused on exploring real-time prediction of epidemic risks at the street and community level in urban scenarios.It contained two main modules:region functional feature learning and region correlation mining.Firstly,the region functional feature learning module integrated the distribution of points of interest(POI)with epidemic risk to express the impact of different urban functions.Secondly,the region correlation mining module maped lower-level grids into higher-level functional regions based on classifications.The proposed model was validated on the Nanjing 2022 COVID-19 dataset.Compared with traditional time series methods and spatio-temporal sequence methods,the proposed method achieved a reduction of 8%-23%in mean absolute percentage error and a reduction of 0.3-1.2 in root mean square error.Additionally,the proposed model also significantly reduced the learnable parameter number.