基于注意力机制的交织区轨迹预测及应用

打开文本图片集

主题词：智能交通 注意力机制 交织区 轨迹预测 冲突检测中图分类号：U491 文献标志码：A DOI：10.19620/j.cnki.1000-3703.20250291

TrajectoryPrediction of Intertwining ZoneBased on Attention Mechanism and Applications

YangHaifei，TangYong，LiuZhe，LiHongwei （Collegeof Civil and Transportation Engineering，Hohai University，Nanjing 210024）

【Abstract】Aiming atthe dificult problemof trajectorypredictionand conflict detectioncausedbyintensive vehiclelane changingin interweavingareasof highways，aSequence-to-Sequence （Seq2Seq）trajectoryprediction model incorporating the attentionmechanismisproposed.Thelocalspatial featuresofvehiclesareextractedbyconvolutionalneuralnetwork，the temporal sequencedependencyis handled bylongandshort-term memorynetwork，and thekeyspatio-temporalinteraction informationisdynamicallycapturedbytheattention mechanism.Basedon the NGSIMdataset，theexperimentalresultsshow that： comparing with theclasical frontiertrajectory prediction model，the maximum reductionofroot mean square error， averageabsoluteerrorandaveragerelativeerroronthehorizontalandverticalcoordinatesofthepredictedtrajectoryof the present model can be up to 97.4%， 97.9%， 99.5% and 77.4% 91.5% ，99.5%，respectively;the improved oriented bounding box algorithmforconflictdetectionoptimizes thedetectionprocess throughtheatentionmechanism，anditsexecutioeficiencyis improvedby6O.2%，whichprovides technicalsupportforthesafetydecision-makingof intellgentvehiclesincomplex scenarios.

KeyWords：Intelligent transportation，Attention mechanism， Interweaving area， Trajectory prediction， Conflict detection

【引用格式】杨海飞，唐勇，刘喆，等.基于注意力机制的交织区轨迹预测及应用[J].汽车技术，2025（12）：1-9. YANGHF，TANG Y，LIU Z，etal.TrajectoryPredictionof Intertwining ZoneBasedonAttention Mechanism and Applications[J].Automobile Technology，2025（12）： 1-9.

1前言

交织区是交通系统中的关键区域，特殊的交通流特征和复杂的驾驶行为使其成为交通拥堵和事故的高发区域。（剩余13197字）