Figures (4)  Tables (4)

    • Figure 1. 

      Flowchart showing the present study.

    • Figure 2. 

      Presentation of the three scenarios. (a) Basic highway segments. (b) Highway merge/diverge areas. (c) Urban expressway weaving zones.

    • Figure 3. 

      Multi-head relative attention mechanism.

    • Figure 4. 

      Convolution-enhanced feedforward neural network.

    • Road scenario Training set Test set Total
      Basic highway segments 601 257 858
      Highway merge/diverge areas 451 193 644
      Urban expressway weaving zones 246 105 351

      Table 1. 

      The number of vehicles in each road scenario.

    • Models Highway basic segments Highway merge/
      diverge areas      		 Urban expressway weaving zones
      RNN 2.15 2.40 2.86
      LSTM 2.11 2.37 2.83
      GNN 2.12 2.34 2.75
      Transformer 2.08 2.35 2.73
      KF-Transformer 1.93 2.07 2.33

      Table 2. 

      Performance of various deep learning models in vehicle trajectory prediction across different scenarios.

    • Models Time (ms)
      RNN 22
      LSTM 37
      GNN 65
      Transformer 71
      KF-Transformer 75

      Table 3. 

      Efficiency of vehicle trajectory prediction by various deep learning models.

    • Models Highway basic
      segments      		 Highway merge/
      diverge areas      		 Urban expressway
      weaving zones
      Transformer 2.08 2.35 2.73
      Transformer-multi-head relative attention mechanism 1.95 2.11 2.45
      Transformer-convolution enhanced feedforward neural network 1.97 2.16 2.40
      KF-Transformer 1.93 2.07 2.33

      Table 4. 

      Ablation experiments of the KF-Transformer for vehicle trajectory prediction.