基于自然驾驶数据的驾驶换道风格分析与辨识  

Analysis and identification of driving lane changing style based on natural driving data

作　　者：刘志强[1] 马进 LIU Zhiqiang;MA Jin(School of Automotive and Traffic Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]江苏大学汽车与交通工程学院,江苏镇江212013

出　　处：《电子设计工程》2024年第9期20-26,共7页Electronic Design Engineering

基　　金：国家自然科学基金(72001095)。

摘　　要：为了研究在换道工况下个体驾驶行为及其辨识,基于NGSIM轨迹数据提取532组换道有效数据。分析处理驾驶员换道特征参数,运用K-means算法对驾驶员进行聚类分析,可分为谨慎型、一般型、激进型三类。并对聚类的参数进行统计分析和时频分析验证各类驾驶人的差异性,并提取出有效参数作为辨识模型的输入,基于惯性权重非线性变化和粒子最大速度非线性递减的策略来改进粒子群,建立起基于改进粒子群优化SVM的驾驶人风格辨识模型,对SVM模型的两个关键参数C和g进行优化更新,驾驶人识别准确率可达到96.25%,且与BP神经网络、PSO-SVM、随机森林、KNN算法相比具有更高的识别准确率。In order to study individual driving behavior and its identification under lane changing conditions,532 sets of effective lane changing data were extracted based on NGSIM trajectory data.The characteristic parameters of drivers’lane changing were analyzed and processed,and the K-means algorithm was used to cluster the drivers,which could be divided into three types:cautious type,general type and aggressive type.And statistical analysis of the parameters of clustering and the time-frequency analysis to verify the differences of all kinds of drivers,and extract the effective input parameters as the identification model,and then based on the nonlinear change of inertia weight and maximum velocity of particles nonlinear regressive strategy to improve the particle swarm,built up based on improved particle swarm optimization driver style identification model of SVM,two key parameters C and g of SVM model were optimized and updated,and the final driver recognition accuracy reached 96.25%.Compared with BP neural network,PSO-SVM,random forest,KNN algorithm,it has higher recognition accuracy.

关 键 词：交通工程 驾驶员行为 改进粒子群优化SVM模型 换道风格识别 辨识模型 

分 类 号：TN01[电子电信—物理电子学]