基于周期性分量提取的城市快速路交通流短时预测理论与方法研究

发布时间：2018-02-01 02:36

本文关键词： 短时预测周期性分量提取傅里叶级数交通流长短时记忆神经网络　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着居民出行总量和机动车保有量的持续增长,城市路网承担的出行压力越来越重。城市快速路作为城市路网的骨架,承载大量的较高车速和长距离出行服务。可靠的短时交通流预测作为交通诱导系统的核心内容,是出行路径诱导与管理者监测控制的技术前提,有助于提高城市路网效率。目前交通流短时预测研究多将交通流量作为一个整体进行预测,没有实现周期性部分和剩余残差项的分离,难以较好地满足预测精度的要求。本文在现有研究成果的基础上,考虑城市快速路交通流的周期性特征,提出基于周期性分量提取的经典模型与长短时记忆神经网络模型,并分别对城市快速路主路和出入口断面流量进行实证研究,研究不同预测步长下周期性分量提取对原始预测模型的改进效果。本文的主要研究内容如下:(1)基于城市快速路交通流动态数据进行数据处理并分析其交通特征,包括交通流数据预处理、交通流参数特征以及皮尔逊相关系数检验。首先针对故障数据进行识别和修复,并进行数据整合等预处理,为后期研究提供可靠的数据基础。其次从宏观交通流参数、参数两两关系和交通流特点等宏观方面分析交通流数据特征。再次根据交通流量存在的日相关性特征进行单一断面流量的Pearson相关系数分析,验证断面流量数据存在周期性。(2)根据工作日期间流量的周期性,结合傅里叶级数理论,建立基于周期性分量提取的城市快速路交通流短时预测模型。首先将流量分为周期性部分和剩余残差项两部分,其次对周期性部分使用傅里叶级数展开,根据傅里叶级数展开项数的不同表示出周期性分量。同时,对逐步分离周期性分量的残差项分别应用自回归积分滑动平均模型、支持向量机回归模型、长短时记忆神经网络模型进行下一时刻残差项的预测,最后与周期性分量合并,实现预测精度的提高。(3)运用基于周期性分量提取的预测模型进行实证性研究。选取北京市二环路四个主路断面以及右安门桥至白纸坊桥路段辅路式独立出入口分别作为快速路主路和出入口研究对象,应用基于周期性分量提取的自回归积分滑动平均模型、支持向量机模型、长短时记忆神经网络预测模型进行单步预测和多步预测。本文的研究成果可为交通流诱导系统提供一种基础方法,并为城市快速路入口控制策略方案的制定提供理论根据。
[Abstract]:With the continuous growth of residents' travel and vehicle ownership, the urban road network bears more and more heavy travel pressure. The urban expressway is the skeleton of the urban road network. Reliable short-term traffic flow prediction, as the core content of the traffic guidance system, is the technical premise of route guidance and manager monitoring and control. It is helpful to improve the efficiency of urban road network. At present, most of the research on short-term traffic flow forecasting takes traffic flow as a whole and does not realize the separation of periodic parts and residual items. It is difficult to meet the requirement of prediction accuracy. Based on the existing research results, the periodic characteristics of urban expressway traffic flow are considered in this paper. A classical model based on periodic component extraction and a long and short time memory neural network model are proposed, and an empirical study on the flow of the main road and the entrance and exit section of the urban expressway is carried out. The effect of periodic component extraction on the original prediction model under different prediction steps is studied. The main contents of this paper are as follows: 1). Based on the urban expressway traffic flow data processing and analysis of its traffic characteristics. Including traffic flow data preprocessing, traffic flow parameters and Pearson correlation coefficient test. First, fault data identification and repair, and data integration and other pretreatment. To provide a reliable data base for the later study. Secondly, from the macro traffic flow parameters. The characteristics of traffic flow data are analyzed in macroscopic aspects such as the relationship between two parameters and the characteristics of traffic flow. Thirdly, the Pearson correlation coefficient of single cross-section flow is analyzed according to the daily correlation characteristics of traffic flow. Verify that the section flow data has periodicity.) according to the periodicity of the flow during the working day, combined with the Fourier series theory. A short-term forecasting model of urban expressway traffic flow based on periodic component extraction is established. Firstly, the traffic flow is divided into two parts: periodic part and residual part, and then Fourier series expansion is used for the periodic part. The periodic components are expressed according to the different terms of Fourier series expansion. At the same time, the autoregressive integral sliding average model and support vector machine regression model are applied to the residual terms of the stepwise separation of periodic components. Long and short time memory neural network model is used to predict the next time residual term, and finally it is combined with the periodic component. To improve the accuracy of prediction. The prediction model based on periodic component extraction is applied to the empirical study. The four main sections of the second Ring Road in Beijing and the auxiliary road entrance of the section of Youanmen Bridge to Baijifang Bridge are selected as the main road and the exit of the expressway respectively. Entrance research object. The support vector machine (SVM) model is applied to the autoregressive integral moving average model based on the periodic component extraction. The prediction model of long and short time memory neural network is used for single step prediction and multi step prediction. The research results in this paper can provide a basic method for traffic flow guidance system. And provides the theoretical basis for the urban expressway entrance control strategy plan formulation.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491

【参考文献】