基于SocialGAN網(wǎng)絡與自注意力機制的車輛軌跡預測方法

打開文本圖片集

主題詞：智能車輛軌跡預測生成式對抗網(wǎng)絡社交池化機制自注意力機制中圖分類號：TP183；U461.91 文獻標志碼：A DOI： 10.19620/j.cnki.1000-3703.20240867

Vehicle Trajectory Prediction Method Based on Social GAN Network and Self-Attention Mechanism

ZhuLangqian’，Ma Shijun'，Liu Mingjian12，LiMuyang'，Hao Changsheng1 （1.Collegeof Information Engineering，DalianOcean University，Dalian16023;2.KeyLaboratoryofEnvironment ControlledAquaculture，Ministryof Education，DalianOcean University，Dalian116023）

【Abstract】Toaddresstheissue thattemporalfeaturesand spatialfactors ofthe trafcenvironment affcttheaccuracyof vehicletrajectorypredictioninvehicledriving，thispaperproposesavehicletrajectorypredictionmethodintegratingtemporal multi-headself-attentionandsocial pooling basedonthe Social GenerativeAdversarial Network （SMA-GAN）.Firstly，the historicaltrajectoryfeaturesareextractedbythetemporalcorelationofthetargetvehicle'sowntrajectorydatausingthemultiheadself-atentionmechanism.Then，thespatialdimensionalfeaturesofthetargetvehicleareextractedbythesocialpoling mechanismbasedonthespatialpositionalrelationshipbetweenthetargetvehicleandthesuroundingvehicles.Finalythe predictedtrajectoryofthetargetvehicleisobtainedbytheencoder-decoder.Modeltrainingandcomparisontestsare conducted using the NGSIM dataset，and theresultsshow that theSMA-GAN model has higher predictionaccuracy and efficiency in the highway scene.

KeyWords：Intelligent vehicle，Trajectory prediction，Generative Adversarial Network（GAN）. Social pooling mechanism，Self-attention mechanism

【引用格式】祝朗千，馬時俊，劉明劍，等.基于Social GAN網(wǎng)絡與自注意力機制的車輛軌跡預測方法[J].汽車技術，2025（6）：8-14.ZHULQ，MASJ，LIUMJ，etal.VehicleTrajectoryPredictionMethodBasedonSocialGANNetworkandSelf-AtentionMechanism[J]. Automobile Technology，2025（6）： 8-14.

1前言

隨著智能交通系統(tǒng)的迅速發(fā)展，車輛軌跡預測技術能夠感知復雜交通環(huán)境中的潛在風險因素，結合駕駛輔助系統(tǒng)的應用，對于提升行車安全具有重要意義。（剩余11141字）