摘要
为实现人机共驾模式下智能系统对驾驶人换道决策的准确识别,将换道决策细分并提出了基于改进的极端梯度提升(XGBoost)的换道决策识别模型。以实车试验采集的自然驾驶数据作为输入,并采用滑动时间窗法确定识别时刻,建立各识别时间窗口下基于XGBoost的换道决策识别模型,同时运用交叉检验和网格搜索(GS)算法进一步提升模型性能,最后利用验证集数据评估所构建GS-XGBoost模型的识别性能,并与机器学习及深度学习模型进行对比。结果表明,所提出的模型在具体换道决策辨识上具有较好的实时性和准确性,且在1.8 s和1.6 s时间窗下的识别准确率最高,达到86.2%。
For the accurate recognition of drivers’lane changing decisions by the intelligent system in man-machine shared driving mode,this paper divides lane changing decisions in detail and proposes an improved eXtreme Gradient Boosting(XGBoost)model for lane changing decision recognition.The identification time is determined by sliding time window method.Taking the natural driving data from vehicle test as inputs,a lane change decision recognition model based on XGBoost is established under each recognition time window.At the same time,cross-check and Grid Search(GS)algorithm are used to further improve the model performance.Finally,the recognition performance of the GS-XGBoost models is evaluated using the validation set data,and compared with the machine learning and deep learning models.The results show that the proposed model has good timeliness and high accuracy in specific lane change decision identification,and the recognition accuracy reaches 86.2%at 1.8 s and 1.6 s time windows.
作者
蒋司杨
李朝
雷毅
王畅
Jiang Siyang;Li Zhao;Lei Yi;Wang Chang(Chang’an University,Xi’an 710064)
出处
《汽车技术》
CSCD
北大核心
2022年第1期27-34,共8页
Automobile Technology
基金
国家重点研发计划项目(2019YFB1600500)。
关键词
人机共驾
换道决策
极端梯度提升
网格搜索
Man-machine shared mode
Lane-changing decision
XGBoost
Grid search
