【作者】 刘翔；

【导师】 成卫； 雷建明；

【作者基本信息】 昆明理工大学 ， 交通运输工程（专业学位）， 2021， 硕士

【摘要】 城市中发生的突发事故通常对城市交通的正常秩序有着非常大的破坏作用,而在突发事故发生后,若应急车辆救援不及时,会导致更大的人员损失与经济损失。因此,应急车辆安全而快速的到达事故现场显得十分重要。虽然法律赋予了应急车辆在道路上的绝对优先通行权,但是社会车辆对应急车辆的通行影响是不可忽略的因素,特别是在交通拥堵路段,应急车辆很难保证优先通行。而城市交通大部分的延误产生于道路交叉口,在交叉口实施应急车辆信号优先控制是保障应急救援的重要手段之一,然而在交叉口实施信号优先控制容易导致非优先方向道路发生严重的拥堵,所以在确保应急车辆在交叉口优先安全通行的同时保证道路交通网络中安全平稳的交通流是非常有必要的。本文针对现有应急车辆信号优先控制算法和模型的不足之处,提出了一种深度强化学习算法来解决多辆应急车辆同时需要信号优先控制的问题,主要内容如下:(1)首先详细阐述了目前已有的应急车辆信号优先控制策略和公交信号优先控制策略,并分析了其存在的不足之处。并结合应急车辆信号优先控制理论,分析了强化学习、深度学习以及深度强化学习应用在应急车辆信号优先控制领域的可行性。(2)其次,目前大部分孤立交叉口应急车辆信号优先控制的研究都针对应急车辆从交叉口某一进口道接近,然后实行应急车辆信号优先控制,而当多辆应急车辆在交叉口不同进口道寻求信号优先时,该方法将不再适用。针对这一问题,本文将应急车辆信号优先控制过程建模为马尔科夫决策过程,通过对有社会车辆与应急车辆混行的交通状态进行编码、重新定义了奖励函数,并利用基于深度双Q学习的方法实现多应急车辆的信号优先控制。通过SUMO仿真实验结果表明,在低流量时段和高流量时段所提出的方法都取得了较好的效果且具有较强的鲁棒性。(3)最后,针对目前基于路线下信号优先控制的不足,通过将孤立交叉口的应急车辆信号优先控制扩展为多路口的应急车辆信号优先控制,提出了采用多智能体A2C算法来解决多辆应急车辆同时在路网行驶的信号优先控制问题。仿真实验表明,在维持应急车辆平均速度和社会车辆平均速度的稳定性方面对比感应优先有着明显优势。更多还原

【Abstract】 The unexpected accidents in the city usually have a very big destructive effect on the normal order of urban traffic,and after the accident,if the emergency vehicle rescue is not timely,it will lead to greater personnel and economic losses.Therefore,it is very important for the emergency vehicle to arrive at the scene of the accident safely and quickly.Although the law gives the emergency vehicles absolute priority on the road,the impact of social vehicles on the emergency vehicles can not be ignored,especially in the traffic congestion section,it is difficult to guarantee the priority of emergency vehicles.Most of the urban traffic delays are caused by road intersections.The implementation of emergency vehicle priority control at intersections is one of the important means to ensure emergency rescue.However,the implementation of priority signal control at intersections can easily lead to the formation of blocking of non priority roads,Therefore,it is very necessary to ensure the safe and stable traffic flow in the road traffic network while ensuring the priority of emergency vehicles at the intersection.In this paper,a deep reinforcement learning algorithm is proposed to solve the problem that multiple emergency vehicles need priority signal control at the same time,The main contents are as follows:(1)Firstly,the existing emergency vehicle priority control strategy and bus priority control strategy are described in detail,and their shortcomings are analyzed.Combined with the emergency vehicle priority control theory,the feasibility of reinforcement learning,deep learning and deep reinforcement learning in the field of emergency vehicle signal priority control is analyzed.(2)Secondly,most of the researches on priority signal control of emergency vehicles at Isolated Intersections focus on emergency vehicles approaching the intersection from an entrance road,and then implement priority signal control of emergency vehicles.When multiple emergency vehicles seek priority signal at different entrance roads of the intersection,this method will not be applicable.To solve this problem,this paper models the priority control process of emergency vehicles as Markov process,codes the traffic state with mixed traffic of social vehicles and emergency vehicles,redefines the reward function,and realizes the priority control of multiple emergency vehicles by using the method based on deep double Q learning.Through sumo simulation results,the proposed method has achieved good results and strong robustness in low traffic period and high traffic period.(3)Finally,aiming at the deficiency of signal priority control based on route,by extending the priority signal control of emergency vehicles at isolated intersections to that of emergency vehicles at multiple intersections,a multi-agent deep reinforcement learning algorithm is proposed to solve the priority control problem of multiple emergency vehicles driving on the road at the same time.The simulation results show that compared with induction priority,induction priority has obvious advantages in maintaining the stability of the average speed of emergency vehicles and social vehicles.更多还原