中文核心期刊
CSCD来源期刊
中国科技核心期刊
RCCSE中国核心学术期刊
重庆交通大学学报（自然科学版）

重庆交通大学学报（自然科学版） ›› 2024, Vol. 43 ›› Issue (2): 84-91.DOI: 10.3969/j.issn.1674-0696.2024.02.11

• 交通+大数据人工智能 • 上一篇    

基于匝道合流数据的自动驾驶汽车安全性测试评价方法

李文礼1,2，李超1，李中峰1，易帆1，李安1   

  1. (1. 重庆理工大学 汽车零部件先进制造技术教育部重点实验室，重庆 400054； 2. 重庆长安汽车股份有限公司，重庆 400020)
  • 收稿日期:2022-12-07 修回日期:2023-09-12 发布日期:2024-03-01
  • 作者简介:李文礼(1983—)，男，河南驻马店人，副教授，博士，主要从事智能汽车测试方面的研究。E-mail:liwenli@cqut.edu.cn
  • 基金资助:
    重庆市自然科学基金面上项目（cstc2021jcyj-msxmX0183）；重庆市高校创新研究群体项目（CXQT21027）；重庆市教委科学技术研究项目(KJQN202201170)

Safety Test and Evaluation Method for Autonomous Vehicles Based on Ramp Merging Data

LI Wenli1,2, LI Chao1, LI Zhongfeng1, YI Fan1, LI An1   

  1. (1.Key Laboratory of Automotive Parts Advanced Manufacturing Technology, Ministry of Education, Chongqing University of Technology, Chongqing 400054, China;2.Chongqing Changan Automobile Co., Ltd., Chongqing 400020, China)
  • Received:2022-12-07 Revised:2023-09-12 Published:2024-03-01

PDF

摘要： 针对自动驾驶汽车测试场景不明确、评价模型主观性强等问题，研究了高速匝道汇入场景下的典型测试场景提取方法和自动驾驶汽车匝道汇入安全性客观评价方法。深入分析了匝道汇入功能场景下的逻辑场景要素，对自然驾驶数据中的自车速度、车间距离、前车车速等逻辑场景要素进行聚类，得到两类典型的匝道汇入测试场景用于自动驾驶汽车的仿真测试。构建多层次自动驾驶汽车评价体系，引入基于自然驾驶数据的核密度估计模型来获取指标最优阈值，建立以最优阈值为参考序列、以层次分析法（AHP）和客观赋权法（CRITIC）为权重输入的灰色关联理评价模型，对自动驾驶汽车在汇入过程中的安全性进行客观评价。评价结果表明：基于核密度估计的灰色关联理论模型评价结果与主观模糊综合分析模型的评价结果相似率达98.01%，验证了客观模型的有效性。

关键词: 车辆工程；测试评价；客观评价模型；聚类分析；核密度估计

Abstract: Aiming at the problems such as unclear test scenarios of autonomous vehicle and strong subjectivity of evaluation models, the extraction method of typical test scene and objective evaluation method of autonomous vehicle on-ramp import safety under high-speed on-ramp import scenario were studied. The logical scene elements in ramp import function scenario were analyzed in depth. The logical scene elements in natural driving data, such as speed of the ego vehicle, distance between vehicles and front vehicle speed were clustered, and two typical ramp import test scenarios were obtained to carry out the simulation test of autonomous vehicles. At the same time, a multi-level autonomous vehicle evaluation system was constructed. A kernel density estimation model based on natural driving data was introduced to obtain the optimal threshold of indicators, and a gray relevance theory evaluation model was established with the optimal threshold as the reference sequence and the analytic hierarchy process (AHP) and objective weighting method (CRITIC) as the weight input. Objective evaluation of the safety of autonomous vehicle during the import process was carried out. The evaluation results show that the similarity between the evaluation results of grey correlation theory model based on kernel density estimation and the evaluation results of subjective fuzzy comprehensive analysis model is 98.01%, which verifies the effectiveness of the objective model.

Key words: vehicle engineering; test and evaluation; objective evaluation model; cluster analysis; kernel density estimation

中图分类号: