【作者】 喻丹；

【导师】 吴义虎；

【作者基本信息】 长沙理工大学 ， 交通信息工程及控制， 2011， 博士

【摘要】 驾驶人接收的信息越来越复杂,感知、判断、操纵行为等失误的概率越来越大,车辆操纵的稳定性也越来越低。同时驾驶人数量剧增、结构参差不齐,行车过程中可靠性在下降。这些因素使得道路交通事故发生率居高不下、死伤人数逐年增长。因此,驾驶人作为交通系统的调节者和控制者,对其行为的研究已成为道路交通安全学的重要研究内容。本文从驾驶人行为建模和可靠性分析两方面开展研究。1.驾驶人行为模型跟驰驾驶行为协同仿真模型研究。以动态交通信息为输入,跟驰车辆响应速度和加速度为输出,设计单个神经元仿真驾驶人对信息的感知筛选过程,采用模糊径向基高斯神经网络提取跟驰行为特征向量作为驾驶经验,应用模糊积分法模拟驾驶人对信息的分析和决策过程,建立跟驰驾驶行为的协同仿真模型模拟跟驰行为中驾驶人信息感知、分析决策和车辆控制的全过程。机动车驾驶人超车行为辨识与预测研究。将超车行为看成为换道行为和切入行为的组合,把超车行为辨识与预测问题转化为换道与切入工况的组合辨识与预测问题。因驾驶人行为受反映交通环境特征的认知信息因子和反映驾驶人特性的主观信息因子共同影响,故模型以本车道前后车速差、本车道前后前间距、相邻左车道前后车速差、相邻左车道前后间距、相邻右车道前后车速差以及相邻右车道前后间距等认知信息因子,注视次数、累积注视时间以及平均眼动速度等主观信息因子为二维输入,以超车行为发生的概率为输出,构建驾驶人超车行为的二维隐马尔科夫辨识和预测模型,组合二维Viterbi算法及log-likelihood评价设计模型求解算法。控制响应延迟下的追尾风险评估。假设驾驶人感知、判定和操作控制决策不存在失误,研究驾驶人车辆控制动作响应延迟对追尾事故的影响。首先定义汽车驾驶人不响应概率函数,采集驾驶人反应时间序列样本,获取驾驶人控制动作的不响应概率;然后基于ANFIS理论,以前后车速差、后车车速、行车间距和驾驶人不响应概率为输入,汽车追尾概率为输出,基于模糊系统与神经网络构建自适应模糊神经系统(ANFIS)追尾风险评估模型。2.驾驶人可靠性分析驾驶人可靠性量化方法研究。从驾驶人行为致因理论出发,将模糊相似测度矩阵代替传统PCA法的协方差矩阵,为驾驶人可靠性增长、维持、衰退三阶段分别筛选影响因子,通过定义驾驶人“可信度”为任意时刻下驾驶人正常完成车辆操作行为的概率量化驾驶人瞬态可靠性并利用由驾驶人反应时间推算所得的差错率对其进行求解,驾驶人整体可靠性由“可信度”的均值和方差共同评价,另外驾驶人可靠性影响因子变化对可靠性的影响也进行了分析。驾驶人可靠性预测方法研究。以后车车速、前后车速差和车间距作为观察变量输入,驾驶人可靠性作为隐含变量输出,构建了机动车驾驶人隐马尔科夫可靠性预测模型。模型通过求解预测时刻所有选中的观察状态序列出现的概率以及各观察状态序列和指定驾驶可靠性状态(隐含变量)同时出现的概率推算驾驶人可靠性处于低水平的概率,并予以报警提示。除考核该预测方法的预报正确性外,还定义了指标“预报度”来衡量驾驶人低可靠性状态概率为P时该方法能实现的提前预警时间,以检验预测方法的预警性能。此外,对各种驾驶人模型开展算例研究,说明模型的有效性;同时指出了需进一步研究的问题。更多还原

【Abstract】 Considering the complexity of traffic information, the probability of drivers’failure has an increment, e.g. drivers’perception, judgment, manipulation, etc., and the vehicle stability control is also becoming less and less. Meanwhile, due to the increasing number and the varying structure of drivers, the reliability of drives during driving is declined. These factors make the high incidence of road traffic accidents, and the number of casualties growing year by year. Therefore, drivers should be the controller of transport system, motor vehicle driver behavior research of road traffic safety has become an important research content. In this paper, the research is based on the driver behavior model and reliability analysis.First, driver’s behavior model is researched:A coordinated simulation model of car-following driving is presented. The whole process of driver behavior from the collecting information, analyzing situation and making decisions to controlling vehicle states are coordinated consider. In which, the dynamic traffic information is input; the velocity and acceleration are output. A single nerve cell is used to simulate how the drivers apperceive the changeable information, a fuzzy never network is imposed to extract the eigenvectors of driving behavior as drivers experience, and the fuzzy integral method is applied to describe the way that drivers analyze the information and make decisions.The identification model and precision model of overtaking behavior is discussed, according to 2-demension Hidden Markov Model. In which, the overtaking behavior is considered as the combination of lane change behavior and cut in behavior. So to identify and precise the overtaking behavior is equal to identify and precise the probability of lane change behavior and cut in behavior in time sequence. Considering neither the cognition information which reflect the traffic environment and the subjective information which reflect the driver character could be dispensed, the input of overtaking model is two-dimension. The cognition information includes the velocity difference with the leading vehicle in the objective lane and the difference distance with the leading vehicle in the objective lane. The subjective information is eye movement parameters, including offixation, gaze duration and average eye movement speed. The probability of Overtaking behavior is output. And the Viterbi algorithm and log-likelihood are used together as the solution of the model. Estimation method of rear-ends accidents caused by the delay of controlling behaviors is studied. Supposed the perception of traffic information and the controlling decision of vehicles are correct. The delay of controlling actions is researched in the paper, and the function of driver non-response probability is advanced. Based on the function, the timing curve of divers’response is made. Then the risk model of Rear-ends accidents is established to evaluate the probability of Rear-end accident, according to the theory named ANFIS, whose input are the headway distance, velocity of the following vehicle, the difference of the leading and following vehicle and the non-response probability, whose output is the probability of Rear-end accidents. And the theory on ANFIS could combine the Fuzzy Theory with the Neural Network, learn from their strong points and close the gap. Second, driver reliability is analyzed:Quantitative method about driver reliability is investigated. The influence factors of reliability for drivers in different running stage are analyzed in the paper, firstly, based on Behavior-causing Theory. Then a new quantification method on transience reliability for drivers is advanced by a new definition“confidence degree”, whose error rate is calculated by the response time of drivers, and the influence factors how to influent the transience reliability degree is also researched. Based on Hidden Markov Model, a new prediction method on driving reliability is advanced. In which, the velocity of following car, the velocity difference and distance headway is input as observation variables, the driver reliability is output as hidden variable. First the probability of observation states needed and the probability of observation states and driver reliability appeared together is forecasted. Then we could get the prediction value of driver reliability, and give some advices. The warning character of the prediction method could be evaluated not only by accuracy but also by a new index‘predictability advanced’, which could show the degree of warning time at p probability.Furthermore, all driver models are applied to analyze their impacts. Finally, the future research emphasis is designated.更多还原