“车适应人”线控汽车驾驶员行为特性辨识算法研究

发布时间：2018-04-30 20:35

本文选题：车适应人 + 驾驶员行为特性　；参考：《吉林大学》2015年硕士论文

【摘要】：传统汽车控制系统由于受到汽车机械结构的限制,汽车的潜能不能很好发挥。随着车载网络、微处理器等技术的迅速发展,很多研究机构和汽车厂家都将线控技术运用到了汽车上。汽车线控技术减少了液压、机械控制装置等部件,降低了整车质量,方便了电线布置,由于线控系统控制算法的灵活多变和系统参数可调,线控汽车的动力学控制比传统汽车具有更大的发展空间,线控汽车已成为国内外研究热点。将驾驶员的特性考虑到车辆集成控制的设计中,就可以实现人性化设计,变“人适应车”的现状为“车适应人”。另外,汽车上各种电控系统的应用及电控系统的集成控制,提高了汽车电子化和智能化水平,在汽车的主动安全性和驾驶舒适性方面发挥了越来越重要的作用。为了提高不同驾驶员对驾驶辅助系统的接受度,在设计控制算法时需要考虑驾驶员的特性,在保证汽车安全性和驾驶舒适性的前提下,通过对驾驶员特性的辨识实现汽车对驾驶员的自适应和驾驶员个性化驾驶。因此无论是要实现“车适应人”线控汽车,还是提高驾驶员对驾驶辅助系统的接受度,都需要对驾驶员特性进行辨识。本文依托于吉林大学汽车仿真与控制国家重点实验室开放基金项目‘'Integrated Control Method for a Full Drive-by-Wire Electric Vehicle Based on Driver's Intention Recognition"(基金编号：20120111)、中国博士后科学基金资助项目(项目编号：2014M561289)和国家自然科学基金青年基金资助项目(项目编号：51305190),以建立一套适用于线控汽车底盘动力学控制系统的驾驶员行为特性辨识算法为目标,搭建驾驶模拟器并选用多名驾驶员进行试验,从采集到的试验数据中提取特征参数并利用K-means进行聚类,获得驾驶员行为特性的类型和数据样本,以此为基础,建立了基于神经网络的驾驶员行为特性辨识模型,并通过驾驶模拟器试验对模型的精度和预测能力进行了验证。为了建立驾驶员行为特性辨识模型,本文主要进行了以下工作： (1)搭建驾驶模拟器为了充分挖掘驾驶员的行为特性,需要进行不同工况下的大量实验,由于实车的可调参数少,以及受环境限制所能选取的工况有限,因此,本文在课题组已有研究基础上,搭建驾驶模拟器进行试验,获取数据样本,为建立驾驶员行为特性辨识模型做准备。本文在提出驾驶模拟器的总体框架和工作原理基础上,详细介绍了驾驶模拟器主体、模拟器操控台、车辆动力学仿真模型、实时仿真系统、转向力感模拟系统和传感器系统等关键组成部分。 (2)对驾驶员行为特性进行分类本文总结并分析了驾驶员行为特性分类的方法,选用K-means算法对驾驶员行为特性进行分类；基于已搭建的驾驶模拟器,设计了转向、制动、加速试验工况,选用13名试验人员进行试验并采集数据；通过对驾驶员转向、制动、加速行为进行分析,选取表征驾驶员转向、制动、加速行为特性的特征参数,并利用MATLAB编程从试验数据中提取特征参数；基于K-means算法对特征参数进行聚类,进而将驾驶员的转向、制动、加速行为特性分别分为谨慎型、一般型和激进型,同时获得每个类型的数据样本,为搭建驾驶员行为特性辨识模型提供数据。 (3)建立驾驶员行为特性辨识模型由于驾驶员特性辨识就是对驾驶员特性这一模式进行识别的过程,因此本文介绍了几种常用的模式识别方法并分析对比了它们各自的优缺点,以及这些模式识别方法用于驾驶员行为特性辨识时的优缺点和适用范围,确定选取BP神经网络作为驾驶员行为特性辨识模型的建模方法。本文利用驾驶员行为分类中所得到的各个类型的数据样本,建立了基于BP神经网络的驾驶员行为特性辨识模型,针对BP神经网络输入输出变量的选取、网络结构的设计以及网络的训练过程进行了详细说明,最后利用驾驶模拟器试验对模型精度及预测能力进行了验证。
[Abstract]:As the traditional automobile control system is limited by the mechanical structure of automobile, the potential of automobile can not be exerted very well. With the rapid development of the technology of vehicle network and microprocessor, many research institutions and automobile manufacturers have applied the wire control technology to the car. The automobile wire control technology has reduced the hydraulic and mechanical control devices and so on. The quality of the whole vehicle is convenient for the layout of the wire. Because of the flexibility of the control algorithm and the adjustable parameters of the system, the dynamic control of the line control car has a greater development space than the traditional car. The line control car has become a hot spot of research at home and abroad. In addition, the application of various electronic control systems on the car and the integrated control of the electronic control system have improved the electronic and intelligent level of the automobile, and played a more and more important role in the vehicle's active safety and driving comfort. The acceptance of the driving auxiliary system needs to consider the driver's characteristics in the design of the control algorithm. On the premise of ensuring the safety and driving comfort of the car, the driver's self adaptation and driver's individualized driving are realized by the identification of the driver's characteristics. To improve driver's acceptance of driver assistance system, we need to identify driver's characteristics.
This paper is based on the open fund project of the State Key Laboratory of automobile simulation and control of Jilin University, "'Integrated Control Method for a Full Drive-by-Wire Electric Vehicle Based on Driver's" (fund number: 20120111), China Post Doctoral Science Fund funded project (project number:) and country The project (project number: 51305190) of the National Natural Science Fund Youth Fund (project number: 51305190) aims to establish a set of driver behavior identification algorithms suitable for the dynamic control system of the car chassis, build a driving simulator and choose a number of drivers to carry out the test, extract the characteristic parameters from the collected test data and use the K-me Ans is used to cluster, and the type of driver behavior and data samples are obtained. Based on this, the identification model of driver behavior based on neural network is established, and the accuracy and prediction ability of the model is verified by driving simulator test.
In order to establish driver behavior characteristic identification model, the following work is done in this paper.
(1) build driving simulator
In order to fully excavate the behavior characteristics of the driver, it is necessary to carry out a large number of experiments under different working conditions, because the adjustable parameters of the real car are few, and the working conditions can be limited by the environment restriction. Therefore, on the basis of the existing research group, this paper builds a driving simulator to test and obtain the data samples, in order to establish the identification of the driver's behavior characteristics. On the basis of the overall framework and working principle of driving simulator, this paper introduces the key components of driving simulator, simulator console, vehicle dynamics simulation model, real time simulation system, steering sense simulation system and sensor system.
(2) classify the behavior of the driver
This paper summarizes and analyzes the method of driver behavior classification, and uses K-means algorithm to classify the behavior characteristics of drivers. Based on the built driving simulator, the steering, braking, acceleration test conditions are designed, 13 experimenters are selected to test and collect data, and the driver's steering, braking, and acceleration behavior are carried out. Analysis, select characteristic parameters that characterizing driver's steering, braking, and acceleration behavior, and use MATLAB programming to extract characteristic parameters from the test data. Based on K-means algorithm, the characteristic parameters are clustered, and then the driver's steering, braking, and accelerating behavior characteristics are divided into discreet, general and radical, and each of them is obtained at the same time. The data samples provide data for building driver behavior identification model.
(3) establish the identification model of driver's behavior
As driver characteristic identification is the process of identifying the driver's characteristics, this paper introduces several common pattern recognition methods and analyzes their respective advantages and disadvantages, and the advantages and disadvantages of these pattern recognition methods used in the identification of drivers' behavior characteristics and the selection of BP neural network. As a modeling method for the identification model of driver's behavior characteristics, this paper sets up a driver behavior identification model based on BP neural network based on all types of data samples obtained in the driver's behavior classification. It aims at the selection of input and output variables of the BP neural network, the design of network structure and the training process of the network. The driving simulator test is used to verify the accuracy and prediction ability of the model.

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.25;U463.6

【参考文献】