基于包络熵的双层分解流感预测模型研究  

A two-layer decomposition model based on envelope entropy for influenza forecasting

作　　者：秦全德[1] 黄兆荣 周至昊 范璧 余乐安 QIN Quande;HUANG Zhaorong;Zhou Zhihao;FAN Bi;Yu Lean(College of Management,Shenzhen University,Shenzhen 518060,China;Business School,Sichuan University,Chengdu 610065,China)

机构地区：[1]深圳大学管理学院,深圳518060 [2]四川大学商学院,成都610065

出　　处：《系统工程理论与实践》2023年第12期3505-3518,共14页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(72174124,71701136);国家自然科学基金重点项目(72331007);广东省基础与应用基础研究基金(2022A1515011009)。

摘　　要：在病毒变异和气候变化等多种内外因素的驱动下,流感的大流行呈现一定的季节性、准周期性和混沌性.在传统多时间尺度分析的基础上,提出了一种基于包络熵和变分模态分解(VMD)的双层分解流感预测模型,将流感复杂时间序列分解成波动相对平缓的有限准周期分量,以降低流感预测的难度.首先,采用VMD方法对流感时序数据进行第一层分解提取主要模态分量特征;然后,在一次分解基础上,通过包络熵来判断复杂度可以减小的分量进行二次分解,以更好地处理复杂时序中较高的非线性和非平稳性特征.最后,结合相空间重构和天牛须算法优化的Elman神经网络实现对双层分量的预测并加总集成.在预测框架中包络熵具备双重作用,既用于优化变分模态分解参数,又用于判断双层分解的合理性.通过南北方省份类流感数据集的实证检验,提出的模型在水平精度和方向精度优于单层分解模型、组合双层分解模型,以及传统单一模型.该流感预测方法为公共卫生部门的流感防控决策提供了一种新的参考.The influenza pandemic exhibits seasonality,quasi-periodicity,and chaos as a result of many internal and external causes such as virus variation and climatic change.Due to these factors,influenza forecasting is complex and challenging.Here we present a new two-layer decomposition forecasting model on the basis of multi-time scale analysis.The time series is decomposed into finite quasi-periodic components with relatively smooth fluctuations to reduce the complexity.In the first layer,the variational mode decomposition(VMD)method is applied to extract the main feature components.In the second layer,the components with potential forecasting power are determined by the envelope entropy for further decomposition into the less nonlinear and stationarity features.Finally,the Elman neural network optimized by phase space reconstruction and Beetle Antennae search algorithm predicts subcomponents in the twolayer.The final prediction result is obtained by summation.The envelope entropy method has dual functions in the proposed framework.One is used to optimize the parameters of variational mode decomposition and the other is to judge the rationality of secondary decomposition.Experimental results of northern and southern provinces’influenza data sets show that the horizontal and directional accuracy of the proposed model is better than that of the single model and other benchmark multiscale prediction models.The proposed method provides a reference for decision-making on influenza prevention in health departments.

关 键 词：流感预测 双层分解 包络熵 变分模态分解 

分 类 号：G353.1[文化科学—情报学]