基于图信号处理的传染病传播预测方法

Prediction method of infectious disease transmission based on graph signal processing

针对现有传染病传播预测模型存在未充分考虑数据的内在关联性的问题,采用图多项式-向量自回归(GPVAR)模型对传染病的传播进行预测,并提出新的用于模型参数估计的优化方法.将传染病发病地区建模为图节点,并根据地区间的距离信息和人群流动情况确定节点间的边及其权重,以反映传染病传播过程中的空间关联性.将不同时刻的感染疾病人数建模为时变图信号,使用GP-VAR模型对时变图信号在图上的演变过程进行预测,并设计一种最小二乘(LS)优化方法对GP-VAR模型的参数进行估计.仿真实验结果表明,与现有的预测方法相比,所提方法能够更好地考虑到数据在空间维的相关性和时间维的演变特性,更加准确地刻画传染病的传播特性,且具有普适性,预测效果更好.

The existing spread predicting models of infectious diseases have not sufficiently considered the intrinsic correlation of the data.To solve this problem,graph polynomial-vector autoregressive(GP-VAR)model was used to predict the spread of infectious diseases,and a new optimization method for estimating model parameters was proposed.The regions where infectious diseases occur were modeled as nodes on the graph,and edges and weights between nodes were determined by the distance information of the regions and the flow of people,so as to reflect the spatial relevance in the transmission process of infectious diseases.The number of cases at different times was modeled as the time-varying graph signal,the GP-VAR model was used to predict the evolution process of the timevarying graph signal on the graph,and a least squares(LS)optimization method was designed to estimate the parameters of the GP-VAR model.Experimental results show that the proposed method can better consider the correlation of data in spatial dimension and the evolution of data in time dimension,and characterize the transmission characteristics of infectious diseases more accurately,which has universality and better prediction effect compared with the existing prediction methods.