基于GRU网络的仿优秀驾驶员换道轨迹模型研究  

Research on Imitation Excellent Driver Lane Trajectory Model Based on GRU Network

作　　者：汪小渟 刘长玉 王俊彦 蔡骏宇[4] 李永博 WANG Xiaoting;LIU Changyu;WANG Junyan;CAI Junyu;LI Yongbo(StarGIS(Tianjin)Technology Development Co.,Ltd.Tianjin 300384,China;Capital Urban Planning&Design Consulting Development Co.,Ltd.Beijing 100000,China;Zhenjiang College,Jiangsu Zhenjiang 212028,China;Jiangsu University,Jiangsu Zhenjiang 212013,China;Tianjin Connected Intelligent Transportation Technology Co.,Ltd.Tianjin 300051,China)

机构地区：[1]星际空间(天津)科技发展有限公司,天津300384 [2]北京市首都规划设计工程咨询开发有限公司,北京100000 [3]镇江高专,江苏镇江212028 [4]江苏大学,江苏镇江212013 [5]天津市网联智能交通技术有限公司,天津300051

出　　处：《交通工程》2024年第7期24-29,43,共7页Journal of Transportation Engineering

基　　金：国家自然科学基金(51675235)。

摘　　要：换道是智能汽车在道路上行驶时必不可少的操作。传统换道模型大多基于数学公式或者车辆动力学、运动学模型,从而忽略驾驶员在实际驾驶过程中的感知与决策能力,生成的轨迹与实际换道轨迹区别很大。本文提出基于GRU网络的仿优秀驾驶员换道轨迹预测模型,利用处理时间序列数据有独特的能力,从优秀驾驶员自身的驾驶特性出发,结合车辆动力学参数,结果表明GRU模型的MAPE为0.044,分别比LSTM和Bi-LSTM降低了24.14%和29.03%;RMSE为0.0111,分别比LSTM和Bi-LSTM降低了0.89%和9.76%,且GRU模型的一致性指数D值更接近1。该模型预测出的换道轨迹与实际轨迹基本一致,能准确地模拟优秀驾驶员的换道轨迹,精度较高,保证安全性的同时兼顾舒适性。Changing lanes is an essential operation for smart cars to perform on the road.The traditional lane changing models,which rely heavily on mathematical formulas or vehicle dynamics and kinematics models,overlook the driver's perceptual and decision-making capabilities during actual driving,resulting in trajectories that differ significantly from real-world lane-changing patterns.The proposed lane-changing trajectory prediction model,based on the GRU network,leverages its unique ability to process time series data.Starting from the driving characteristics of excellent drivers and incorporating vehicle dynamics parameters,the model demonstrates a high degree of accuracy.The results indicate that the GRU model achieves a MAPE of 0.044,representing a reduction of 24.14%and 29.03%compared to LSTM and Bi-LSTM,respectively.The RMSE is 0.0111,which is 0.89%and 9.76%lower than LSTM and Bi-LSTM,respectively.Moreover,the consistency index D value of the GRU model is closer to 1.The predicted lane-changing trajectories align closely with actual trajectories,accurately simulating the lane-changing behavior of excellent drivers.This model not only ensures safety but also prioritizes comfort.

关 键 词：换道轨迹 优秀驾驶员模型 深度学习 GRU网络 

分 类 号：U491.2[交通运输工程—交通运输规划与管理]