交通拥堵区域的发现与预测技术研究

发布时间：2018-02-26 03:16

本文关键词： 出租车GPS数据交通拥堵基于距离聚类马尔可夫链预测　出处：《哈尔滨工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：从全国乃至全世界的交通情况来看,随着各个国家的机动车数目的不断增加,现有的公共交通条件将越来越不能达到机动车容量对其的要求,从而将会导致各种交通系统的问题,现如今交通拥堵这个问题的普遍存在已经是人们和社会都不能忽视的一个严峻问题,而解决此问题最好的办法便是预防,预防交通拥堵问题的出现并将交通拥堵扼杀在摇篮中。另外,本文通过分析交通系统中所存在的不足得知ITS系统中存在数据资源浪费等问题。针对交通拥堵问题以及智能交通系统的数据资源浪费问题,本文首先利用智能交通系统所采集到的时空数据找出交通拥堵区域,其次根据交通拥堵区域的发现结果预测各个区域在之后的时间出现交通拥堵情况的几率。本文利用的是智能交通系统中所采集到的数据,选取了北京市出租车系统中的12,000辆出租车于2012年11月所反馈的GPS数据作为数据源。本文首先根据GPS定位系统原理所造成的数据噪声以及本文实情将数据集进行了清洗及时间片分割的操作。其次本文根据拥堵区域车辆密集的特点针对时间片数据集进行基于距离聚类分析(K-means、DBSCAN聚类算法),并且将两种聚类方法的结果及性能作比较,本文最终根据比较结果的分析选取了DBSCAN聚类算法来分析各个时间片的数据集。时间片数据集在进行聚类后,将算法得到聚类结果与分割后的网格区域相匹配,并将区域分为聚类数据簇内部、聚类数据簇边缘、聚类数据簇外部。然后本文将各个区域的车辆平均时速进行计算,其中将聚类数据簇外部车辆时速记为零。本文根据拥堵区域判定规则得到每个时间片上各个区域的交通拥堵情况,并将交通情况细分为“严重拥堵”、“中度拥堵”、“轻度拥堵”、“畅通”,最后将最终结果以矩阵的形式存储于文本文件当中。由于导致交通拥堵的原因较为复杂多变,所以出现交通拥堵的情况有一定的随机性。交通拥堵状况可以看做是当前时刻的状态只依赖于上一个时刻的状态,所以本文根据马尔可夫链链预测模型建立交通拥堵情况预测模型,将交通拥堵情况的发现结果分为训练集和验证集。其中利用训练集进行基于Markov链的交通拥堵预测,利用验证集来验证该模型的正确率。最后,本文将K-means、DBSCAN聚类、基于Markov链预测模型作了实验及对比,并对预测模型正确率进行验证统计。
[Abstract]:From the point of view of the traffic situation of the whole country and even the whole world, with the increasing number of motor vehicles in each country, the existing public transportation conditions will be more and more unable to meet the requirements for the capacity of motor vehicles. This will lead to a variety of traffic system problems. Nowadays, the prevalence of traffic congestion is a serious problem that people and society cannot ignore, and the best way to solve this problem is to prevent it. Prevent traffic congestion and stifle it in its cradle. In addition, By analyzing the shortcomings of traffic system, this paper finds out that there are some problems in ITS system, such as waste of data resources, traffic congestion problem and data resource waste problem of intelligent transportation system. In this paper, we first use the space-time data collected by the Intelligent Transportation system to find out the traffic congestion area. Secondly, the probability of traffic congestion in each area is predicted according to the results of the traffic congestion area. This paper uses the data collected in the intelligent transportation system. The GPS data of 12,000 taxis in Beijing taxi system on November 2012 is selected as the data source. Firstly, the data set is carried out according to the data noise caused by the principle of GPS positioning system and the fact of this paper. Secondly, based on the characteristics of vehicle density in congested areas, this paper makes a distance based clustering analysis of time slice data sets, and compares the results and performance of the two clustering methods. Finally, DBSCAN clustering algorithm is selected to analyze the data sets of each time slice according to the analysis of comparison results. After clustering, the clustering result is matched with the segmented grid area. The region is divided into cluster data cluster interior, cluster data cluster edge, cluster data cluster outside. Then, the average speed of vehicles in each region is calculated. In this paper, the traffic congestion of each region on each time slice is obtained according to the decision rules of congestion area. The traffic situation is subdivided into "severe congestion", "moderate congestion", "mild congestion", "smooth flow", and the final result is stored in a text file in matrix form. So traffic jams have a certain randomness. Traffic jams can be seen as the state of the current moment that only depends on the state of the previous moment. In this paper, the traffic congestion prediction model is established according to Markov chain forecasting model, and the traffic congestion detection results are divided into training set and verification set, in which traffic congestion prediction based on Markov chain is carried out by using training set. Finally, the K-means-DBSCAN clustering, based on the Markov chain prediction model, is tested and compared, and the accuracy of the prediction model is verified and counted.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491;O211.62

【参考文献】