结合可视图的多状态交通流时间序列特性分析邢雪于德新田秀娟王世广吉林大学交通学院吉林化工学院信息与控制工程学篋吉林大学吉林省道路交通重点实验室摘要：交通流时间序列的研宄主要采用数据挖掘和机器学习的方法，这些“黑箱”挖掘方法很难直观反映序列特性.为增强交通流时间序列及其特征分析的可视化性，结合可视图理论来构建交通流吋间序列的关联网络，从复杂网络角度实现交通流时间序列的特性分析.在网络构建的过程屮，考虑到不同交通状态卜交通流表征具有的差昇性，首先利用交通流参量的相关性对交通流状态进行分类，然后构建不同交通状态下的时间序列复杂网络，并对这些网络的特征属性给出统计分析，如度分布、聚类系数、网络直径、模块化等.研究表明，可视图法可为交通流吋间序列映射到网络提供有效途径，并且不同状态下交通流吋间序列构建的复杂网络的模块化、聚类系数和度分布等统计特征呈现一定的变化规律，为交通流运行态势的研究提供了可视化的分析角度.关键词：交通流吋间序列;可视图;复杂网络;聚类分析;作者简介：邢雪，E-mail:patricia_xx@126.com收稿日期：2017年3月19日基金：W家自然科学基金（批准号:51308248）Analysisofmulti-statetrafficflowtimeseriespropertiesusingvisibilitygraphXingXueYuDe-XinTianXiu-JuanWangShi—GuangCollegeofTransportation，JilinUniversity;Abstract：Atrafficflowtimeseriesisasequenceoftrafficdetectionparametersinchronologicalorder.Thisdiffersfromageneralquantitativedatasequenceinthatthetimeseriesincludesatimeattributethatcontainsnotonlythedatawithtimecharacteristics，butalsothedistributionofthedataitself.Todate,studiesoftraffictimeserieshaveprimarilyadopteddataminingmethodsconsistingofdataminingandmachinelearningmethods—similarsequencesearch,dimensionreduction,clustering,classification,patternanalysis,prediction,etc.Inordertoimprovethevisualizationoftrafficflowtimeseriesandfeatureanalyses,aproposedmethodbuildstheassociationnetworksoftrafficflowtimeseriesbyusingvisibilitygraphtheory.Thisapproachdiffersfromtraditionaltrafficflowtheoryasitperformsfeatureanalysisoftrafficflowtimeseriesfromtheperspectiveofcomplexnetworks,andthenanalyzestherelationshipbetweenthecharacteristicsofthestructureinthevisualnetworkandthestatecharacteristicsofthetrafficflow.Theproposedmethodalsotakesintoaccountthedifferenttrafficflowtimesequencesthatcorrespondtodifferenttrafficstates.Inthenetworkbuildingprocessusingtheproposedmethod，thetrafficflowisclassifiedbycorrelatingthetrafficflowparameterstothestructureofthecomplextimeseriesnetworksunderdifferenttrafficconditionsthroughconsideringthechangesintrafficflowcharacteristicsundervarioustrafficconditions.Next，statisticalanalysesofthesignsandattributesofthenetworks(e.g.degreedistribution,clusteringcoefficient,networkdiameter,andmodularization)areconducted.Theanalysisresultsshowthattheproposedvisibilitygraphmethodcanprovideaneffectiveapproachtomappingtrafficflowtimeseriestothenetwork.Moreover,themodularity,clusteringcoefficient,anddegreedistributionofthetrafficflowtimeseriesnetworksindifferenttrafficstatesshowspecificallyvaryingpatterns,providingawaytovisuallyanalyzethetrendsintrafficflowoperation.Whenthetrafficconditionisatlevel1,thedistributionofthescatteredpointsofthenetworkconformstoapowerlawdistribution.Whenthetrafficconditionisatanyotherlevel,thedistributionofthescatteredpointsofthenetworkisconsistentwithaGaussiandistribution.Themodularityofthetimeseriesnetworkalsoshowssomestatisticalcharacteristics,thatis,thenumberofmodulesgrowsrapidlywhenthetrafficstateswitchesfromsmoothtomoderatecongesti...