基于电动出租车数据的充电桩选址聚类方法比较.pdfVIP

基于电动出租车数据的充电桩选址聚类方法比较

摘要：为有效降低出租车运营企业及经营者的经济成本，通过分析出租车的卫星轨迹数据，比较和选取用于电动出租

车充电桩选址规划的聚类方法。以上海市电动出租车充电站的选址规划为研究对象，分别基于孤立森林和聚类算法设计

异常值检测方法，对相关时段的出租车卫星数据进行清理以及数据可视化处理；比较层次聚类（Agglomerative

Clustering）、高斯混合模型（GaussianMixtureModel，GMM）、K-means聚类、Mean-Shift聚类以及谱聚类（Spectral

Clustering）5种算法的聚类效果，并选取K-means算法作为充电桩选址规划参考算法。从城市区域划分及企业运营角度

确定充电桩选址方案，为未来上海市区电动出租车充电桩的数量和容量配置提供设计依据。

关键词：电动出租车；充电桩选址；异常值检测；聚类方法；可视化

ComparisonofClusteringMethodsforChargingStationSite

SelectionBasedonElectricalTaxiData

Abstract:Inordertoeffectivelyreducetheeconomicexpensesfortaxicompaniesandusers,theanalysisof

satellitetrajectorydataforelectricaltaxisisusedtocompareandselectsuitableclusteringmethodsfor

chargingpilelocationplanning.Focusingonthelocationplanningofelectricaltaxichargingstationsin

Shanghai,thepaperdesignedtheoutlierdetectionmethodbasedonisolatedforestandclusteringalgorithms

tocleanupthetaxisatellitedatawithintherelevanttimeperiod,followedbydatavisualizationprocessing.

Theclusteringeffectsoffivealgorithms,includingAgglomerativehierarchicalclustering,GMMGaussian

mixedclustering,K-meansclustering,Mean-shiftclusteringandSpectrumclusteringwereevaluatedand

compared.AndtheK-meansalgorithmwasselectedasthereferencealgorithmforchargingpilelocation

planning.Fromtheperspectivesofurbanzoningandbusinessoperations,asiteselectionstrategyis

determined,whichprovidesafoundationforthedesignandplanningofthequantityandcapacityallocation

ofelectrictaxichargingpilesinShanghaiforthefuture.

Keywords:electricaltaxi;chargingpilesiteselection;outlierdetection;clusteringmethods;visualization

收稿日期：2022‒06‒16改稿日期：2022‒09‒13

第4期