信号控制交叉口交通拥堵状态识别方法研究

发布时间：2018-06-18 14:07

本文选题：交通拥堵 + 信号控制　；参考：《华南理工大学》2015年硕士论文

【摘要】：交通拥堵已经成为世界各国高度关注和亟待解决的问题。由交通拥堵导致的交通能耗、环境污染,是我国城市面临的极其严重的“城市病”之一。交通拥堵识别可以起到预防和缓解城市交通拥堵的作用,因此对其研究具有重要意义。针对城市道路交通拥堵特征,结合信号控制交叉口的特点,本文主要研究讨论了基于贝叶斯决策的信号控制交叉口拥堵识别方法,贝叶斯训练样本集的更新方法,基于朴素贝叶斯的信号控制交叉口交通拥堵状态识别系统的设计。对于信号控制交叉口拥堵识别方法的研究,以贝叶斯基础理论和朴素贝叶斯分类器模型为基础,本文提出了一种基于朴素贝叶斯决策的信号控制交叉口拥堵识别方法,该方法把交通拥堵的识别看作是一个不确定性的分类问题,把交通状态分成畅通、拥挤和拥堵三种状态,将交通流量、占有率和排队长度比作为判别参数,通过学习在畅通、拥挤和拥堵三种状态下的历史数据,生成贝叶斯分类器,然后利用分类器对实时采集到的数据进行分类,从而识别交通状态。贝叶斯分类器模型是以历史训练样本的概率表来决策的,所以最佳的训练样本自然能够决定分类器最优的分类趋向。训练样本的准备是贝叶斯分类的基础工作,但在实际中很难获取完备的训练样本,且仅仅依靠一成不变的历史数据来识别交通状态,全面性也不够的。基于此,本文提出了一种增量学习方法来更新训练样本,并对识别算法进行改进,即按照一定的规则将分类好的数据添加到训练集中,动态更新训练集,丰富训练信息,使得拥堵的识别结果更加可靠。通过VISSIM仿真获取的数据对贝叶斯算法进行了分析和评价,算法的误判率为6.92%,表明算法对信号控制交叉口的拥堵识别具有可行性和实用性。最后,以理论引导实践,介绍了信号控制交叉口交通状态识别系统的总体架构,构建了基于朴素贝叶斯方法的信号控制交叉口交通拥堵状态识别系统,包括功能设计和数据库设计。在C#编程语言集成开发环境下,采用结构化的设计思想,通过软件实现所阐述的功能,实现交通状态的识别。
[Abstract]:Traffic congestion has become a serious concern and urgent problem in the world. Traffic energy consumption and environmental pollution caused by traffic congestion are one of the most serious "urban diseases" faced by cities in China. Traffic congestion identification plays an important role in preventing and alleviating urban traffic congestion, so it is of great significance to study it. According to the traffic congestion characteristics of urban roads and the characteristics of signal-controlled intersections, this paper mainly discusses the congestion identification method of signal-controlled intersections based on Bayesian decision, and the updating method of Bayesian training sample set. Design of signal-controlled intersection traffic congestion recognition system based on naive Bayes. On the basis of Bayesian theory and naive Bayesian classifier model, this paper presents a new method of congestion identification for signal-controlled intersection based on naive Bayesian decision. In this method, traffic congestion identification is regarded as an uncertain classification problem, traffic state is divided into three states: unblocked, congested and congested, traffic flow, occupation rate and queue length ratio are taken as discriminant parameters, and the traffic flow, occupancy ratio and queue length ratio are used as discriminant parameters. The historical data of congestion and congestion are used to generate Bayesian classifier and then the real-time collected data are classified by classifier to identify the traffic state. Bayesian classifier model is based on the probability table of historical training samples, so the best training samples can naturally determine the optimal classification trend of classifiers. The preparation of training samples is the basic work of Bayesian classification, but it is very difficult to obtain complete training samples in practice, and only rely on historical data to identify traffic state, comprehensive is not enough. Based on this, an incremental learning method is proposed to update the training samples, and the recognition algorithm is improved, that is, the classified data is added to the training set according to certain rules, the training set is updated dynamically, and the training information is enriched. Make the identification of congestion more reliable. The Bayesian algorithm is analyzed and evaluated by the data obtained by VISSIM simulation. The error rate of the algorithm is 6.92. it shows that the algorithm is feasible and practical to identify the congestion of signal-controlled intersection. Finally, the paper introduces the overall framework of signal-controlled intersection traffic state recognition system, and constructs a signal-controlled intersection traffic congestion recognition system based on naive Bayesian method. Including function design and database design. In the C # programming language integrated development environment, the structure design idea is adopted, and the function described is realized by software to realize the recognition of traffic state.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.23

【参考文献】