融合图卷积网络与社交池化的多模态轨迹预测模型

赵庆慧,崔鑫,张艺炜,陈燕

Multimodal trajectory prediction model integrating graph convolutional networks and social pooling

Qinghui ZHAO,Xin CUI,Yiwei ZHANG,Yan CHEN

表 1 不同轨迹预测模型在NGSIM数据集上的性能对比

Tab.1 Performance comparison of different trajectory prediction models on dataset NGSIM

模型 RMSE/m NLL tp=1 s tp=2 s tp=3 s tp=4 s tp=5 s 平均值 tp=1 s tp=2 s tp=3 s tp=4 s tp=5 s 平均值 CV 0.73 1.78 3.13 4.68 6.68 3.400 3.72 5.37 6.40 7.16 7.76 6.082 M-LSTM 0.58 1.26 2.13 3.23 4.67 2.374 0.87 2.49 3.46 4.09 4.74 3.130 PIP 0.56 1.23 2.10 2.99 4.21 2.218 1.78 3.36 4.24 4.87 5.41 3.932 RA-LSTM 0.52 1.06 2.23 3.16 4.55 2.304 1.46 2.61 3.84 4.49 5.03 3.486 WSiP 0.54 1.17 2.04 3.08 4.14 2.194 1.66 3.30 4.17 4.80 5.32 3.850 HAN 0.52 1.01 1.69 2.52 3.54 1.856 1.36 3.14 4.04 4.68 5.18 3.680 STA-LSTM 0.54 1.20 1.97 2.94 4.17 2.164 0.56 1.57 2.32 3.14 4.04 2.326 HTSA-LSTM 0.49 0.98 1.53 2.34 3.46 1.760 0.51 1.28 2.06 2.77 3.26 1.976 HLTP++ 0.48 0.98 1.56 2.17 3.32 1.702 0.61 1.69 2.54 2.89 3.79 2.304 GC-LSTM 0.47 0.89 1.41 2.08 3.22 1.614 0.48 1.12 1.87 2.47 2.98 1.784