基于XGBoost算法的商用车驾驶风险辨识模型
Commercial Vehicle Driving Risk Recognition Model Based on XGBoost Algorithm
王永亮 1李超 2许恩永 2何水龙3
作者信息
- 1. 桂林电子科技大学 机电工程学院,广西 桂林 541004
- 2. 东风柳州汽车有限公司,广西 柳州 545005
- 3. 桂林电子科技大学 机电工程学院,广西 桂林 541004;东风柳州汽车有限公司,广西 柳州 545005
- 折叠
摘要
驾驶员作为交通事故的重要诱因,其一系列风险驾驶行为对道路交通安全具有重要影响.针对当前驾驶风险等级分类不合理、辨识精度低等问题,提出一种商用车驾驶风险辨识模型.即首先从车辆状态、驾驶状态、驾驶员操作三个方面,共建立18个能够表征商用车驾驶风险的特征参数;采用因子分析法(FA)对特征参数降维优化,并生成蕴含更为明确风险驾驶行为信息的综合变量;接着应用K-means聚类算法分别将风险驾驶行为特征聚为2、3和4类并对比分析,结合肘部法则和轮廓系数综合确定最佳的聚类数目k,消除人为经验确定k值主观性强的缺陷;最后,利用极端梯度提升(XGBoost)算法对商用车驾驶风险进行识别,并与决策树、随机森林、K近邻等算法在精度上进行比较.研究结果表明:在上述研究条件下XGBoost算法对商用车驾驶风险的理论识别率最高可达98%,该结果对于自动驾驶辅助系统的设计、道路交通安全性的提升具有重要意义.
Abstract
Drivers are an important cause of traffic accidents,and a series of risky driving behaviors have an important impact on road traffic safety.Aiming at the current problems of unreasonable classification of driving risk levels and low identification accu-racy,a commercial vehicle driving risk recognition model is proposed.Firstly,from three aspects of vehicle state,driving state and driver operation,18 characteristic parameters that can represent the driving risk of commercial vehicles are established.Fac-tor analysis method(FA)is used to optimize the dimensionality reduction of characteristic parameters,and comprehensive vari-ables containing more explicit information of risky driving behaviors are generated.Then,the K-means clustering algorithm is applied to cluster risky driving behavior characteristics into 2,3 and 4 categories and make a comparative analysis.Combined with elbow rule and contour coefficient,the best number of clusters is determined to eliminate the subjective defect of artificial em-pirical determination of k value.Finally,the Extreme Gradient Boosting(XGBoost)algorithm is used to recognize the driving risk of commercial vehicles,and the accuracy is compared with decision tree,random forest,k-nearest neighbor and other algo-rithms.The research results show that under the above research conditions,XGBoost algorithm has the highest theoretical identi-fication rate of commercial vehicle driving risk,up to 98%,which is of great significance for the design of automatic driving as-sistance system and the improvement of road traffic safety.
关键词
风险驾驶行为/因子分析/K-means聚类/XGBoost算法/驾驶风险辨识/道路交通安全Key words
Risky Driving Behavior/Factor Analysis/K-Means Clustering/XGBoost Algorithm/Recognition of Driving Risk/Road Traffic Safety引用本文复制引用
基金项目
广西创新驱动发展专项(AA18242033)
柳州市科技计划项目(2020GAAA0404)
出版年
2024
NETL
NSTL
万方数据