新冠肺炎传播预测优化:基于少量早期数据的多阶段SEIRr模型

Improving COVID-19 Prediction:A Multi-stage SEIRr Model Using Limited Data from Early Stage

使用传染病传播动力学模型预测重大新发突发传染病时,极易发生因早期统计数据与实际不符以及对传播特征的认识不足等原因而造成参数估计偏差,导致预测、分析结果不准确。本文结合新冠疫情肺炎传播特征改进SEIR模型,提出了考虑无症状感染者、自愈人群以及无效接触人群的多阶段SEIRr模型,并使用批量梯度下降算法进行参数学习,针对不同场景下的多个国家与地区的传播过程进行预测。实验结果表明,SEIRr模型拟合用数据更少,对数据质量要求更低,预测效果更好;而基于有限早期数据的多阶段模型能有效刻画新发突发传染病防控和治疗手段不断优化的过程,具有较好的拟合预测效果和通用性。

Traditional compartment models of infectious diseases are often limited in predicting emerging major contagious diseases with parameter estimation error and inaccurate predictions and analysis,largely due to the lack of reliable epidemiological data and that the knowledge of the transmission mechanism and treatment is unclear in the early stages of the outbreak.In this article,we propose a modified SEIR model based on the transmission mechanism of COVID-19,SEIRr,considering the asymptomatic infections,self-healing populations,and invalid contacts,model the spreading of COVID-19 in different cities and countries.The batch gradient descent algorithm is used to learn the model’s parameters.Experimental results indicate that SEIRr model yields improved prediction with less fitting data and has lower requirements for data quality;the multi-stage model based on limited early data has better prediction performance and universality,which can effectively characterize the evolution of COVID-19 mitigation measurements and treatment options.