高速公路场景下的车辆换道意图预测研究

打开文本图片集

关键词：车辆换道意图识别；图注意力网络；图卷积网络；注意力机制中图分类号：TP391.4 文献标志码：A 文章编号：1001-3695（2025）12-007-3574-08doi：10.19734/j. issn.1001-3695.2025.05.0158

Research on vehicle lane-changing intention prediction in expressway scenarios

Huang Haifeng，Huang Deqi†，Huang Deyi，Liu Zhenhang （School ofElectrical Engineering，XinjiangUniversity，Urumqi 83Oo17，China）

Abstract：Vehicle lane-changing behavironhighways incomplex dynamic scenarios is highly prone to traficaccidents and affcts roadcapacity.Toimprovetheabilitytopredictlane-changing intentions，thisstudyproposedadual-channel GATMGCNmodel integratedwithaself-atentionmechanismforlane-changing intentionrecognition.Themodelcombinedhighorder interactionfeatures extracted byGATv2 with topological structurefeatures captured by GCNthroughalearnablemesgepassing function，fusingtheminhigh-dimensional space.Meanwhile，itemployed linearprojectiontoeliminateredundantinformationandareconstruction mechanismto preserve critical features，significantly enhancing discriminative featurerepresentationwhileoptimizingmodelrobustnessandlightweightdesign.Throughthisjointfeaturerepresentation，themodelffectively integratedspatiotempralinteractioninformationfrommulti-sourceheterogeneousdata，substantiallimprovingtheauracyof lane-changing intentionrecognition.Toaddressdataimbalance，thestudyoptimized themodel usingadynamiclossfunction withclass weights.Evaluationsonthepublic HighDhighwaytrajectorydatasetdemonstrate thatthe modeloutperforms traditional machine learningand existingdeeplearning methods interms of accuracy，precision，andreal-time performance.Ablationexperimentsfurthervalidatethekeycontributionsoftheproposedapproach tomodelperformance.Thisresearchprovides anovel solution for predicting lane-changing intentions in highway scenarios.

Key words：lanechange intention identification；graph atentionnetwork（GATv2）；graph convolutionalnetwork（GCN）; attention mechanism

0 引言

高速公路行驶下，车辆的换道行为普遍存在。（剩余16257字）