摘要
自2019年12月以来,新冠肺炎(COVID-19)疫情在全球范围内持续扩散,不仅严重危害到世界各国人民的生命健康,对公共医疗卫生体系提出严苛考验,还对经济贸易活动造成了巨大冲击,对国际社会产生了深远影响.一些研究采用数学预测模型对病毒传播和疫情发展进行模拟仿真,以帮助研究人员和政策制定者了解病毒传播机理,采取合理防疫政策进而抑制病毒进一步传播.然而现有研究存在一定局限性,例如方法选择单一、过于依赖模型参数选择、病毒传播与政策调整导致的数据时变性等问题.为解决上述问题,本文提出了基于时变模型平均(TVJMA)、时变参数模型(TVP)、传染病vSIR模型(vSIR)、逻辑回归模型(LR)、多项式回归模型(PNR)、自回归移动平均(ARMA)六种模型方法的综合集成预测框架,对不同地区疫情最为严重的6个国家的累计确诊人数进行预测.结果表明,对于单一预测方法,TVJMA方法表现优于其他五种方法;综合集成预测方法在绝大多数情况下明显优于单一方法,特别是基于误差修正权重的多模型组合预测方法,显著地提高了预测精度.对于不同预测步长,综合集成预测方法具有稳健性.
Since December 2019,COVID-19 epidemic is continuing to spread globally.It not only jeopardizing the lives and health of people around the world seriously and putting a severe test on the public medical and health system,but also causes a huge impact on economic and trade activities and has a deep influence on the international community.In order to help researchers and policy makers understand the mechanism of virus transmission and adopt reasonable anti-epidemic policies to inhibit the further spread of the virus,some studies have adopted mathematical prediction models to simulate the spread of the virus and the development of the epidemic.However,the existing research has certain limitations,such as single method selection,excessive reliance on model parameters selection,and virus transmission and policy adjustments caused time variability of data.To solve the above problems,this paper proposes a comprehensive ensemble forecasting framework,which bases on six single prediction models,including time-varying Jackknife model averaging(TVJMA),time-varying parameters(TVP),time-varying parameter SIR(vSIR),logistic regression(LR),polynomial regression(PNR),autoregressive moving average(ARMA).The proposed method is used to predict the cumulative number of confirmed cases in the 6 most severely affected countries in different regions.Empirical results show that for a single prediction method,the TVJMA method outperforms the other five methods;the comprehensive ensemble forecasting method is significantly better than any single method in most cases,especially,the multi-model combined forecasting method based on error correction weights improves the prediction accuracy significantly.For different prediction steps,the comprehensive ensemble forecasting method is robust.
作者
白云
钱箴
孙玉莹
汪寿阳
BAI Yun;QIAN Zhen;SUN Yuying;WANG Shouyang(Academy of Mathematics and Systems Science,Chinese Academy of Sciences,Beijing 100190,China;School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190,China;Center for Forecasting Science,Chinese Academy of Sciences,Beijing 100190,China)
出处
《系统工程理论与实践》
EI
CSSCI
CSCD
北大核心
2022年第6期1678-1693,共16页
Systems Engineering-Theory & Practice
基金
国家自然科学基金面上基金项目(72073126)
国家自然科学基金基础科学中心项目“计量建模与经济政策研究”(71988101)
国家自然科学基金重大项目子课题(72091212)
国家能源集团2021年度十大软课题《能源系统模型构建与中国能源展望研究》(GJNY-21-141)。
关键词
COVID-19
综合集成预测
时变模型平均
非参估计
传染病vSIR
MCS检验
COVID-19
ensemble forecasting method
time-varying Jackknife model averaging
non-parametric estimation
time-varying parameter SIR
MCS test
