基于两次聚类的PWARX驾驶行为辨识模型  

PWARX Driving Behavior Identification Model Based on Twice Clustering

作　　者：应海宁[1] 唐振民[2] 韩旭[2] YING Haining;TANG Zhenmin;HAN Xu(College of Electronic Information Engineering,Nanjing Vocational Institute of Transport Technology,Nanjing 211188,Jiangsu,P.R.China;School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,Jiangsu,P.R.China)

机构地区：[1]南京交通职业技术学院电子信息工程学院,江苏南京211188 [2]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《重庆交通大学学报（自然科学版）》2019年第2期79-85,101,共8页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：国家自然科学基金项目(61305134)

摘　　要：针对传统基于聚类的PWARX模型依赖先验知识和子空间划分精度不高的问题,提出一种基于两次聚类的PWARX辨识模型改进算法,并将其应用于驾驶行为建模。首先通过近邻传播算法在样本空间上进行聚类,在所得类簇上用线性模型进行拟合,并通过K-means算法在线性模型的参数空间上进行聚类,以获得PWA子模型的区域划分,最后在各个子空间上对PWA子模型进行求解。该算法合理利用了近邻传播算法和K-means算法的特点,通过在样本和参数空间上的两次聚类获得了良好的子模型区域分割效果。并对10名驾驶员的驾驶行为的建模实验结果表明,所提算法的模型辨识的平均准确率达到了91.5%。According to the draw backs of traditional clustering-based PWARX model such as the dependence of prior knowledge and the low accuracy of subspace segmentation,an improved algorithm for PWARX identification model based on twice clustering was proposed,which was applied to driving behavior modeling.Firstly,clustering in the sample space was carried out by affinity propagation(AP)algorithm.Then the linear models were used to fit the obtained clusters,and K-means algorithm was used to cluster in the parameter space of the linear model to obtain the region partition of the PWA sub-model.Finally,the PWA sub-models were solved in each subspace.The proposed algorith mrationally took advantage of the characteristics of the AP algorithm and the K-means algorithm,and obtained a good sub-model region segmentation effect by twice clustering in the sample and parameter space.The proposed algorithm was used to model the driving behavior of 10 drivers.The results show that the average accuracy of model identification of the proposed algorithm is 91.5%.

关 键 词：交通工程 PWARX模型 近邻传播聚类 K-MEANS聚类 危险驾驶行为 

分 类 号：U491.255[交通运输工程—交通运输规划与管理] O235[交通运输工程—道路与铁道工程]