摘要
为了在驾驶人避让决策研究中考虑驾驶人认知不确定性,并获得确定性避让决策规则,提出一种基于云模型和粗糙集的无人行横道过街处驾驶人避让决策分析方法。首先,划分距离和速度等避让决策影响因素的分级概念,建立相应的数据序列,采用逆向云发生器提取表征影响因素分级概念的数字特征,再采用基于条件云发生器的隶属概念判别方法实现距离和速度等数据的离散化处理,进而构建了无人行横道过街处驾驶人避让决策表。然后,采用粗糙集理论中基于差别矩阵的属性约简算法和基于归纳的值约简算法,进行决策表约简和无人行横道过街处驾驶人避让决策规则提取;约简后的条件属性为欲到达车辆与行人间的距离、欲到达车辆的速度和行人的速度,提取出23条无人行横道过街处驾驶人避让决策规则,其中包括17条确定性规则和6条不确定性规则,并举例说明确定性规则和不确定性规则的含义。最后,通过预测正确率和ROC曲线下面积2个指标比较,分析该方法与现有研究中的决策树和Logistic回归方法的差异。研究结果表明:所提方法的预测正确率为92.2%,分别比决策树和Logistic回归方法提高3.9%和1.3%;ROC曲线下面积为0.968,分别比2种方法提高了1.1%和3.2%;所提方法的预测性能较好,并能得到简单直观的无人行横道过街处驾驶人避让决策规则,可为交通安全仿真研究奠定基础。
In order to consider driver cognitive uncertainty and obtain certainty yielding decision rules in the study of driver yielding decision, a method for driver yielding decision analysis at unmarked roadway based on cloud model and rough set was proposed. Firstly, the grading concepts of the factors affecting yielding decision such as distance and speed were divided, and the corresponding data sequence were established, then the reverse cloud generator was used to extract the digital features that characterize the grading concepts of the influencing factors, and then the membership concept discriminant method based on conditional cloud generator was used for the discretization of data such as distance and speed, and then the driver yielding decision table at unmarked roadway was established. Then, attribute reduction based on discernibility matrix and value reduction based on induction in the rough set theory were applied to reduce the decision table and extract the driver yielding decision rules at unmarked roadway. After the reduction, the conditional attributes were the distance between the approaching vehicle and the pedestrian, the speed of the approaching vehicle and the speed of the pedestrian. Twenty-three driver yielding decision rules at the unmarked roadway were obtained, including seventeen certainty rules and six uncertainty rules. The meaning of certainty and uncertainty rules was illustrated. Finally, the proposed method was compared with the decision tree and Logistic regression method in the existing research by using the prediction accuracy rate and the area under the ROC curve. The results show that the prediction accuracy of the proposed method is 92.2%, which is higher than those of the two comparison methods by 3.9% and 1.3% respectively, and the area under the ROC curve of the proposed method is 0.968, which is higher than those of the two comparison methods by 1.1% and 3.2% respectively. The proposed method has good prediction performance and can obtain the simple and intuitive driver yieldin
作者
陈鹏
余敬柳
谢静敏
CHEN Peng;YU Jing-liu;XIE Jing-min(School of Transportation,Wuhan University of Technology,Wuhan 430063,Hubei,China)
出处
《长安大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2020年第3期84-90,共7页
Journal of Chang’an University(Natural Science Edition)
基金
国家自然科学基金项目(51208400)。
