基于Attention机制和递归思想的LSTM车辆轨迹预测

LSTM Vehicle Trajectory Prediction Based on Attention Mechanism and Recursion Thought

针对现有车辆轨迹预测模型在长时预测方面准确性不足的问题,基于Attention机制和递归思想的长短时记忆网络(long short-term memory, LSTM)构建了一种新型的车辆轨迹预测模型,即ATT-LSTM(RE)模型,使用编码器–解码器架构更精确地预测车辆未来的行驶轨迹。研究结果表明,模型意图识别的准确率为91.7%,F1分数、召回率、精确率均在0.872~0.977之间;1 s、2 s、3 s、4 s、5 s的终点轨迹预测的均方根误差为0.52 m、1.07 m、1.69 m、2.58 m、3.31 m,优于同类型模型。

The vehicle trajectory prediction models couldn’t accurate enough in long-term forecasting.Based on the integration of the Attention mechanism and the recursion thought Long Short-Term Memory(LSTM)network,a new novel model was developed.An encoder-decoder architecture was employed to enhance the prediction accuracy of future vehicle trajectories.Research revealed that the accuracy of the model intent recognition is 91.7%.The F1 score,recall,and precision range from 0.872 to 0.977.The mean square error for endpoint trajectory predictions at 1 s,2 s,3 s,4 s,and 5 s were 0.52 m,1.07 m,1.69 m,2.58 m,and 3.31 m,respectively.The results were superior to similar models.