采用双向LSTM自编码器的驾驶风格谱聚类识别研究  

Study of spectral clustering using bi-directional LSTM autoencoder for driving style recognition

作　　者：梁科 陈华晟 潘明章 叶宇[2] LIANG Ke;CHEN Huasheng;PAN Mingzhang;YE Yu(The School of Mechanical Engineering,Guangxi University,Nanning 530004,China;Yuchai Engineering Research Institute,Guangxi Yuchai Machinery Co.,Ltd.,Nanning 530007,China)

机构地区：[1]广西大学机械工程学院,南宁530004 [2]广西玉柴机器股份有限公司玉柴工程研究院,南宁530007

出　　处：《重庆理工大学学报（自然科学）》2023年第10期28-37,共10页Journal of Chongqing University of Technology：Natural Science

基　　金：广西科技重大专项(桂科AA22068103);广西重点研发计划(桂科AB201220059);广西科技基地和人才专项(2021AC19324);广西制造系统与先进制造技术重点实验室(20-065-40S006)。

摘　　要：不同驾驶风格的分类对驾驶安全、道路设计和燃油经济性具有深远的影响。考虑到驾驶风格受驾驶员即时操作和前后操作的影响,提出了一种采用双向LSTM自编码器的谱聚类模型对驾驶风格进行识别,以反映驾驶数据时序性对驾驶风格识别的影响。首先利用鲸鱼优化算法对驾驶过程生成的自然驾驶数据进行特征选择,再利用基于双向LSTM的自编码器模型,获得用于谱嵌入的特征值和特征向量,并最终通过谱聚类对驾驶风格进行识别。应用本文中所提出的方法对真实驾驶数据进行比较分析。结果表明:该方法在聚类的精确性优于SOM和LSTM-谱聚类方法。此外,该方法还能在降低数据特征的情况下有效地识别驾驶员的驾驶风格,并反映驾驶员的操作策略。The recognition of different driving styles has profound implications for driving safety,road design and fuel economy.Considering that driving styles are affected by drivers’immediate and back-and-forth operations,this paper proposes a bi-directional LSTM autoencoder-based spectral clustering model for driving style recognition,in order to address the influence of driving data temporality on driving style recognition.Firstly,a whale optimization algorithm is used to select features from the real-time data from the driving process.Secondly,an autoencoder-based bi-directional LSTM model is built to obtain the eigenvalues and eigenvectors for spectral embedding.Finally,the driving styles are recognized by spectral clustering.The analysis of the real-time driving data shows the accuracy of the proposed method is higher than that of SOM and LSTM-based spectral clustering.Besides,the proposed method can effectively identify drivers’driving style and reflect their operating strategies with fewer features.

关 键 词：驾驶风格识别 双向LSTM 自编码器 谱聚类 

分 类 号：U471[机械工程—车辆工程] TP399[交通运输工程—载运工具运用工程]