多源数据驱动的细粒度传染病预测模型  

A Fine-Grained Infectious Disease Prediction Model Driven by Multi-Source Data

作　　者：李锦宇 阮思捷 许皓翔 杜婧[2] 唐易成 LI Jinyu;RUAN Sijie;XU Haoxiang;DU Jing;TANG Yicheng(Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学,北京市100081 [2]北京市疾病预防控制中心,100013

出　　处：《中国卫生信息管理杂志》2024年第5期661-668,687,共9页Chinese Journal of Health Informatics and Management

基　　金：国家重点研发计划“重大传染病传播风险社会数据化治理智能技术研究”(2023YFC2308703);国家自然科学基金“时空大数据驱动的细粒度地理信息收集与城市空间建模方法研究”(62306033)。

摘　　要：目的 研究基于多源数据的传染病细粒度预测模型,为传染病精准防控提供依据。方法 基于传染病历史确诊数据以及来自医疗机构和社会的外部数据,构建多源多粒度时空网络(MMSTNet)。MMSTNet充分融合了不同空间粒度的数据,采用图注意力网络捕捉空间相关性,采用门控循环单元捕捉时间相关性,预测未来细粒度传染病确诊人数。结果 MMSTNet在各预测天数下预测误差均小于基线模型,其平均绝对误差比最佳基线模型误差降低14.4%。结论 融合来自医疗机构和社会的外部数据、考虑区域间的空间相关性,能够有效提升细粒度传染病预测准确性。Objective To develop a fine-grained infectious disease prediction model based on multi-source data,providing a basis for precise prevention and control of infectious diseases.Methods Based on historical confirmed case data of infectious diseases and external data from medical institutions and society,we propose a Multi-source Multi-grained Spatio-temporal Network(MMSTNet).It fully integrates data of different spatial granularity,leverages graph attention networks to capture spatial correlations,and gated recurrent units to capture temporal correlations,and predicts the number of fine-grained confirmed cases of infectious diseases in the future.Results The prediction error of MMSTNet is smaller than all baselines over all prediction days,with its mean absolute error reduced by 14.4%compared to the best baseline.Conclusion Integrating external data from medical institutions and society,and considering spatial correlations between regions,can effectively improve the accuracy of fine-grained infectious disease predictions.

关 键 词：传染病预测 多源数据 时空预测 细粒度建模 图注意力网络 

分 类 号：R-058[医药卫生] R183