Tunnel driving risk identification based on multi-source data
A multi-source data-based method was proposed to identify driving risks in complex tunnel environ-ments.Multi-source driving datasets were obtained based on real-vehicle driving experiments.A sliding time window was used for driving sample construction in the entrance,interior and exit sections of a tunnel.The samples in terms of risk were evaluated by the spectrum of undesirable driving behaviors.A light gradient boosting machine(LGBM)was used to construct a driving risk identification model.The risk influencing fac-tors in different sections were analyzed by the partial dependency diagram(PDP).The results show that the driving risk inside the tunnel is lower,while those in the entrance and exit sections are higher.The areas under the precision-recall curves of LGBM on the test set of the entrance,interior,and exit sections are 0.888,0.893 and 0.860,respectively.The driver and road environment characteristics can effectively improve the performance of the driving risk identification model.The risk influencing factors of different road sections are different.Compared with the inside of the road section,the entrance and exit sections are more significantly affected by a variety of factors.
万方数据
维普
