基于时空关系协同作用的交通数据预处理方法研究

发布时间：2018-01-21 06:12

本文关键词： 数据预处理时空相关性小波分析车道相关性异常识别　出处：《长安大学》2015年硕士论文　论文类型：学位论文

【摘要】：目前数据时代已经慢慢的走进了我们的工作、生活中,这就是大数据平台分析对于城市发展决策带来的直观体现。大数据平台是一个包罗万象的数据仓库,为了使所含数据在被分析过程中尽可能得到准确的结果,所以其数据都需要经过一个清洗与筛选的过程,而交通大数据作为此平台中最主要的一类数据集,其必然也要经过这样的处理过程。道路监测系统监测得到的数据信息对交通项目设计与评价等起着至关重要的作用,其监测统计数据的准确性直接关系到未来决策的方向。目前的道路交通监测系统主要采用检测线圈以及视频监视设备等对道路交通数据进行统计,由于设备的老化,软件的故障等因素常会发生统计结果的错误,进而导致以数据为研究基础的其他分析无法进行或分析结果错误,故交通数据的预处理过程是具有重要意义的。本文首先分析了常见的数据异常性识别方法,并从其适用性及运算可行性的角度分析了各自的优缺点。以此为基础本文从交通数据具有的时间周期特性与车道空间分布特性的角度出发,提出了基于交通数据的时间关系与空间关系协同作用的时空关联预处理的方法。在交通数据时间相关性方面,为了使数据在识别与修复过程中充分依托于数据自身的周期特性,同时数据在识别端具有可变尺度进行调节,故引入了对时间序列数据具有多尺度可变分析的小波变换,并在小波变换理论中以四阶牛顿-科茨定理的合理使用改进了算法,大大减少了计算困难。在交通数据的空间相关性方面,本文从道路中相邻车道的车道利用率角度出发,以相邻车道交通流分布规律性的不同为基础,进行了数据的异常识别,同时以相邻车道空间关系对数据进行修复处理。由信息熵下的误差熵值概念将时间关系与空间关系协同作用,得到更有的时空关系修复结果。最后本文以南宁市良庆区五象大道与金象大道交叉口东进口的三直行车道为研究实例,对中间直行车道的交通量数据进行了异常识别与修复,并以误差分析作为修复效果评价指标,分别对未经处理的数据、只进行小波变换的修复结果、小波变换与相邻车道空间关系协同作用的的修复结果这三个情况进行了误差分析比对,从分析结果上来看在结合了时间相关性的小波变换与空间相关性的相邻车道修复的结果要明显优越于单一处理的修复结果。
[Abstract]:At present, the data age has gradually entered our work, life, this is the big data platform analysis for urban development decision-making intuitive embodiment. Big data platform is an all-encompassing data warehouse. In order to get as accurate results as possible in the process of analysis, the data must go through a process of cleaning and filtering, and traffic big data is the most important data set in this platform. It must also go through such a process. The data information obtained from the road monitoring system plays a vital role in the design and evaluation of traffic projects. The accuracy of monitoring statistical data is directly related to the direction of future decision-making. The current road traffic monitoring system mainly uses detection coil and video surveillance equipment to statistics the road traffic data. Due to the aging of equipment, software failures and other factors will often occur statistical error, which leads to other data-based analysis can not be carried out or the results of the analysis errors. Therefore, the process of traffic data preprocessing is of great significance. Firstly, this paper analyzes the common data anomaly identification methods. And from the point of view of applicability and operational feasibility of the analysis of their advantages and disadvantages, based on the traffic data with the time cycle characteristics and lane spatial distribution characteristics of the point of view. In this paper, a method of spatio-temporal correlation preprocessing based on the synergistic effect of time relation and spatial relation of traffic data is put forward, which is related to the time of traffic data. In order to make the data fully rely on the periodic characteristics of the data in the process of identification and repair, at the same time, the data has variable scale to adjust in the recognition end. Therefore, the wavelet transform with multi-scale variable analysis for time series data is introduced, and the algorithm is improved by the rational use of fourth-order Newton-Coates theorem in wavelet transform theory. In terms of the spatial correlation of traffic data, this paper starts from the perspective of lane utilization of adjacent lanes, based on the traffic flow distribution regularity of adjacent lanes. At the same time, the data is repaired by the spatial relationship of adjacent lanes. The time relationship and spatial relationship are combined by the concept of error entropy under information entropy. Finally, this paper takes the three straight lanes at the east entrance of the intersection of Wuxiang Avenue and Jinxiang Avenue in Liangqing District, Nanning City as an example. In this paper, the traffic volume data of the middle straight lane are identified and repaired, and the error analysis is used as the evaluation index of the repair effect. The unprocessed data are only repaired by wavelet transform. The error analysis and comparison of the results of the synergetic effect of wavelet transform and adjacent lane spatial relationship are carried out. From the analysis results, the result of adjacent lane restoration which combines time correlation wavelet transform and spatial correlation is obviously superior to that of single processing.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491

【相似文献】