Q学习结合神经网络的智能驾驶行为决策系统

发布时间：2018-11-17 11:54

【摘要】：近年来,随着全球经济的持续快速发展,汽车数量急剧增长,与此同时,交通事故的发生率也随之增加。交通事故统计数据表明,人的因素在引发交通事故的因素中占重要地位。如果驾驶员在驾驶决策过程中,能准确感知周围车辆的实时信息,如车辆位置、速度、行驶方向等,就能在第一时间内做出正确的判断和操作,以避免交通事故的发生。然而驾驶员在短时驾驶决策中,信息处理能力是非常有限的,无法同时获取人、车、路、环境等多源信息,以致有时不能准确、快速地判断车辆所处的环境而引发交通事故。因此,智能感知车辆环境信息并处理这些信息,帮助驾驶员判断当前环境下的最优驾驶行为,即给驾驶员提供辅助驾驶决策,对确保准确、快速的驾驶决策和道路交通安全至关重要。目前针对车辆环境感知和驾驶行为决策的研究层出不穷,但是如何保证准确全面的获取车辆信息以及如何降低驾驶行为的决策的时间仍然是富有挑战性的研究方向。针对以往驾驶行为决策算法数据获取存在误差和时延的问题,本文在前期研究基于WAVE协议的车联网组播通信的基础上,实现了车与车(V2V)应用层面的数据快速交互,保证了数据的全面性和准确性。在基于WAVE协议实现了 V2V快速可靠的数据交换后,设计了以多车预警和结合神经网络的Q学习决策为一体的智能决策系统,对数据交换的信息进行处理。该智能决策系统在Repast J仿真环境中进行测试,结果表明本系统能够实现车与车之间快速可靠的通信;且数据获取时间和决策算法的时间都较短,总时延只有0.5s,能为驾驶员提供实时的最优驾驶行为决策,以辅助驾驶员安全驾驶。
[Abstract]:In recent years, with the sustained and rapid development of the global economy, the number of cars has increased sharply, and the incidence of traffic accidents has also increased. Statistics of traffic accidents show that human factors play an important role in causing traffic accidents. If the driver can accurately perceive the real time information of the surrounding vehicle, such as the position, speed and direction of driving, the driver can make correct judgment and operation in the first time, so as to avoid the traffic accident. However, in short time driving decision, the information processing ability of driver is very limited, and he can not get the multi-source information of person, vehicle, road, environment at the same time, so that he can not judge the environment of the vehicle accurately and quickly and cause the traffic accident sometimes. Therefore, intelligently perceiving the vehicle environment information and processing the information to help the driver to judge the optimal driving behavior in the current environment, that is, to provide the driver with auxiliary driving decision, to ensure the accuracy, Fast driving decisions and road safety are crucial. At present, there are many researches on vehicle environment perception and driving behavior decision, but how to obtain vehicle information accurately and comprehensively and how to reduce the decision time of driving behavior is still a challenging research direction. Aiming at the problem of error and delay in data acquisition of previous driving behavior decision algorithms, this paper realizes the fast data interaction between vehicle and vehicle (V2V) application level on the basis of the previous research on vehicle network multicast communication based on WAVE protocol. It ensures the comprehensiveness and accuracy of the data. After realizing V2V fast and reliable data exchange based on WAVE protocol, an intelligent decision system based on multi-vehicle early warning and Q-learning decision combined with neural network is designed to process the information of data exchange. The intelligent decision system is tested in Repast J simulation environment. The results show that the system can realize fast and reliable communication between vehicles. The time of data acquisition and decision algorithm is short, the total delay is only 0.5 s, which can provide the driver with real-time optimal driving behavior decision to assist the driver to drive safely.
【学位授予单位】：天津工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6

【相似文献】