节点文献
城市交通拥堵状态自动判别方法研究
【作者】 王东；
【导师】 陈笑蓉；
【作者基本信息】 贵州大学 ， 计算机应用技术， 2008， 硕士
【摘要】 本论文依托国家自然科学基金资助项目《城市交通若干问题研究》(10671045)和贵阳市科技局基金资助项目《贵阳市道路交通数据库建设和网络模型研究》,研究城市道路交通拥堵状态自动判别方法。通过对城市道路交通拥堵特性的分析,运用分类理论设计了一种用于城市道路的交通拥堵状态自动识别(ACI)算法,该法把交通拥堵是否发生看作一个特殊的分类问题,在不考虑路段受信号灯影响的情况下,把交通状态分成拥堵和畅通两种状态,将交通量、速度、占有率作为交通参数,通过学习在拥堵和畅通两种状态下的历史数据,生成贝叶斯分类器,然后用分类器对实时检测到的交通数据进行分类,从而判别路段交通状态。微观交通仿真数据的实验表明了该方法的可行性和有效性。针对训练集含有噪音样本的问题,若这些样本参与训练学习往往会弱化分类性能,本文提出了训练集增量优化算法,算法将原始训练集分成两个部分,首先以第一部分为基础,增量获取另一部分的较优子集,再以该子集为基础,增量获取第一部分的较优子集,此过程交替迭代、交互验证,最后得到原始训练集的最优子集。该方法不需要设定阈值,优化过程充分利用了样本信息。实验表明经优化训练集学习得到的分类器可以有效提高分类精度。本文最后对城市道路交通拥堵自动判别系统的总体框架及增量贝叶斯交通拥堵判别子系统进行了设计,给出了系统的设计流程。结合提出的增量型贝叶斯交通状态自动判别算法,对判别子系统的功能模块及数据库进行了设计。
【Abstract】 With the supports of the projects "Study on Multiple Problems in Municipal Transportation System" of the Natural Sciences Foundation of China, and "Study on Guiyang Road Traffic Database Construction and Network Modeling" of Guiyang Science and Technology Foundation, this thesis focuses on algorithms for automatic identification of traffic congestion.A method of traffic congestion identification based on Bayes classifier theory is presented through analysis of municipal traffic congestion characteristic. Whether traffic congestion occurs or not is considered as a special classification problem. If the branch road of city is not affected by the signal light, the traffic situation is divided into tow parts: congestion and unimpeded. Using data associated with the traffic parameter for congestion and non congestion, an incremental Bayes classifier is trained to detect whether traffic congestion occur or not. Experimental results based on microcosmic traffic simulation indicate that this method is not only feasible but also effective.To reduce the effects of noise in training data set on the capability of classifiers, an incremental optimize algorithm is proposed, in which primitive training set is firstly spitted into two parts, and the first part is trained to look for the superior subset from the second part. Secondly, the superior subset is trained to obtain a new superior subset from the first part, and iterate this process in turn to obtain the most superior training subset of primitive training set. This method needs not any set threshold, and be convergence automatically. Besides, the optimized process fully uses the sample information. Experiments show that the classification precision can be enhanced by incremental optimize algorithm.Finally, this thesis designed the overall framework of the municipal road traffic automatic identification system and the incremental Bayes traffic congestion identification subsystem, and presented some proposals for the function module design and the database construction.