分簇数据收集的协同分布式Q学习交通信号配时优化

发布时间：2018-05-15 08:59

本文选题：车载网自组织网络 + 分簇数据收集　；参考：《中南大学》2014年硕士论文

【摘要】：随着世界各国城市化进程的加速,城市交通拥堵问题已经成为当今世界许多城市所面临的难题。利用基于车载自组织网络(VANET)收集实时交通数据,对交叉路口信号灯进行配时优化,为用户提供便捷交通引导服务具有重要研究意义。针对VANET网络拓扑快速动态变换和交通信号灯配时优化问题,本文以减少城市拥堵,提高道路利用率为目标,对VANET中分簇交通数据收集和信号灯配时优化这两个关键问题进行研究。首先,以增强网络拓扑稳定性,提高数据传输率,降低通信开销为目标,提出一种动态分簇的交通数据收集算法。为适应VANET网络中车辆节点的动态特性,在车对车通信模式(V2V)下,采用近邻传播簇头选择算法,将邻居节点集、车辆速度、节点间距离和车道权重值作为簇头选择判据,对簇内节点进行评估,建立适应VANET网络的分簇结构；采用车与基础设施通信模式(V2I),簇头节点实时收集交通数据并发送至交叉路口智能体,为交叉路口信号灯进行配时优化提供实时的交通状态信息。其次,针对大规模城市交通系统中车流非连续性、时变性、随机性等特点,提出一种快速梯度下降的协同分布式Q学习信号配时优化算法。建立交通信号配时优化中的Q学习模型,利用VANET网络收集的实时交通数据,对交叉路口各车道车辆排队长度进行估计；通过交换相邻路口的交通状态信息,根据交叉路口间协同行为,设计无需中央监控代理的优化策略。为提高信号配时优化算法的实时性,引入快速梯度下降因子,设计函数逼近方法,解决协同分布式Q学习中动作行为对呈指数增长的维数灾难问题；并对传统Q学习中的ε-贪婪策略进行改进,寻求搜索和利用平衡策略,加快算法收敛速度。利用VanetMobiSim和NS-2对交通数据分簇收集算法联合仿真,使用GLD和MATLAB对交通信号配时优化方案进行仿真,验证论文所提算法的有效性。图27幅,表2个,参考文献71篇。
[Abstract]:With the acceleration of the process of urbanization in the world, the problem of urban traffic congestion has become a difficult problem in many cities in the world. It is of great significance to use the vehicle based auto organization network (VANET) to collect real-time traffic data, optimize the timing of intersection signals and provide convenient traffic guidance services for users. For the fast dynamic transformation of VANET network topology and the optimization of traffic signal timing, this paper aims at reducing urban congestion and improving road utilization, and studies the two key problems of cluster traffic data collection and signal timing optimization in VANET.
First, in order to enhance the network topology stability, improve the data transmission rate and reduce the communication overhead, a dynamic clustering algorithm for traffic data collection is proposed. In order to adapt to the dynamic characteristics of the vehicle node in the VANET network, the neighbor transmission cluster head selection algorithm is adopted under the vehicle to vehicle communication mode (V2V), and the neighbor node set, vehicle speed and node are used. The interval and lane weight value are used as cluster head selection criteria to evaluate the cluster nodes and establish the cluster structure adapted to the VANET network. Using the vehicle and infrastructure communication mode (V2I), the cluster head nodes collect traffic data in real time and send to the intersection agent to provide real-time traffic shape for the intersection signal optimization. State information.
Secondly, in view of the characteristics of discontinuity, time variability and randomness in large-scale urban traffic system, a fast gradient descending cooperative distributed time optimization algorithm for cooperative distributed Q learning signal is proposed. The Q learning model of traffic signal timing optimization is set up, and the real-time traffic data collected by VANET network is used to arrange vehicles in each lane of intersection. In order to improve the real-time performance of the signal timing optimization algorithm, the fast gradient descent factor is introduced and the function approximation method is designed to solve the action behavior of the cooperative distributed Q learning. The dimension disaster problem is exponential growth, and the epsilon greedy strategy in the traditional Q learning is improved to search for and use the balance strategy to speed up the convergence speed of the algorithm.
The traffic data clustering algorithm is simulated by VanetMobiSim and NS-2. The traffic signal timing optimization scheme is simulated by GLD and MATLAB, and the validity of the proposed algorithm is verified. Figure 27, table 2, and 71 references.

【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.54

【参考文献】