融合主成分分析与支持向量机的船舶异常行为识别方法.pdfVIP

融合主成分分析与支持向量机的船舶

异常行为识别方法

摘要：为解决船舶异常行为识别率低下的问题，综合考虑船舶属性和运动特征，提出了一种融合主成分分析和支

持向量机的船舶异常行为识别方法。根据船舶作业过程对船舶行为模式进行划分，并明确船舶行为模式相关的船

舶运动特征；进而以船舶自动识别系统数据为基础，运用主成分分析算法提取最具代表性的船舶运动特征；最后运

用支持向量机算法对船舶异常行为进行识别。选取青岛港附近水域的船舶轨迹进行试验分析，结果表明：青岛港

附近水域最具代表性的船舶运动特征共10个，总贡献率为84.40%；船舶异常行为识别结果的平均精确率和平均

召回率分别为83%和84%，均优于对比模型。研究成果可以为船舶位置异常、轨迹异常、航速异常、航向异常等异

常情况的智能发现和海事监管提供支撑

关键词：船舶异常行为；主成分分析；支持向量机；轨迹特征挖掘

Identifyingshipsabnormalbehaviorbyusingprincipalcomponent

analysisandsupportvectormachineincombination

Abstract:Inordertosolvetheproblemoflowrecognitionrateofabnormalbehaviorofships,amethodofidentifying

abnormalbehaviorofshipsbasedonPrincipalComponentsAnalysis

proposed,whichcomprehensivelyconsidersshipattributesandmotioncharacteristics.Theshipbehaviorpatternsare

dividedbyconsideringtheshipoperationprocess,andtheshipmotioncharacteristicsrelatedtotheshipbehaviorpatterns

aredefined;Then,basedonAutomaticIdentificationSystem

representativeshipmotionfeatures.Finally,SVMalgorithmisusedtoidentifytheabnormalbehaviorofships.Theship

trajectoryinthewatersnearQingdaoPortisselectedforexperimentalanalysis.Theresultsshowthatthereare10most

representativeshipmotionfeaturesinthewatersnearQingdaoPort,withatotalcontributionrateof84.40%.Theaverage

accuracyrateandaveragerecallrateofshipabnormalbehavioridentificationresultsare83%and84%,respectively,

whicharebetterthanthecomparisonmodel.Theresearchresultscanprovidesupportforintelligentdiscoveryandmaritime

supervisionofabnormalsituationssuchasabnormalshipposition,abnormaltrajectory,abnormalspeedandabnormal

course.

Keywords:abnormalshipbehavior;principalcomponentanalysis;supportvectormachine;trajectoryfeaturemining

收稿日期：2022-07-04

陈家豪，等：融支特量利行