突发事件下道路交通流预测方法研究

发布时间：2018-12-17 12:14

【摘要】：随着我国经济的飞速发展,交通需求也日益增加,随之而来的交通事故也大幅提升,由突发事件引起的道路拥堵、环境污染和安全问题,已引起交通管理者和交通领域科研学者的广泛关注。突发事件状态下的交通流具有突变、偶发及非线性的特征,其内部机理和功能之间的复杂性、关联性尚未得到足够的重视。实时准确地对事发路段的交通流进行快速预测不仅有助于提高道路交通管理水平,也有助于人们更合理的规划自身的出行路径和时间。当前在交通流预测领域存在很多预测方法,例如基于时间序列数据的ARIMA预测法、基于空间状态数据的卡尔曼滤波预测法和基于时空数据的支持向量机预测法,这些预测方法一方面,预测流程较复杂,且需要对模型的参数进行多次设置。另一方面,在预测过程中都未考虑事件对交通流预测准确度的影响。针对上述问题,本文从突发事件下道路交通流的时空特性入手,深入分析了事件状态下各相邻路段上检测器交通流数据的时空相关性,提出了一种考虑事件因素的随机森林预测法,最终通过实际数据验证了本方法的有效性。本文的研究成果概括如下：(1)提出了基于122报警信息的交通事件获取方法,建立了北京市地名地址空间位置信息数据库,在此基础上通过“分层分块”的地名数据组织方式,提取相应层和相应区域的地名作为地名词典。同时以交通事件的空间位置信息为基础,将检测器采集的交通流数据进行标识,从而知道每个数据的具体状态。(2)以获取的交通事件为基础,将采集的交通流数据划分为：有事件时的交通流数据和无事件时的交通流数据,通过对事件状态下交通流速度变化规律的研究,发现突发事件状态下相邻检测器的数据具有很强的时空相关性。(3)以交通流的时空相关性和获取的交通事件为基础,考虑突发事件本身的因素,将分类后的数据,分别用时间序列数据和空间状态数据作为随机森林、ARIMA和卡尔曼滤波算法的输入,对突发事件下的交通流进行预测。(4)为更好的提高预测精度,提出了一种数据时空融合的预测方法,该方法主要是借助最小二乘法将时间序列数据预测法和空间序列预测法获得的预测结果进行融合处理,最终获得一组新的预测结果。同时引入三个误差评价指标：误差百分比绝对值均值(MAP)、误差绝对值均值(MA)和误差平方均值(MS),对算法进行评价。
[Abstract]:With the rapid development of our country's economy, the traffic demand is increasing day by day, and the traffic accidents have been greatly increased, resulting in road congestion, environmental pollution and safety problems caused by unexpected events. Traffic management and researchers in the field of transportation have attracted wide attention. The traffic flow in the emergency state has the characteristics of mutation, contingency and nonlinearity, the complexity of its internal mechanism and function, and the correlation has not been paid enough attention to. It is not only helpful to improve the level of road traffic management, but also help people to plan their own travel path and time more reasonably and accurately. At present, there are many forecasting methods in the field of traffic flow prediction, such as ARIMA forecasting method based on time series data, Kalman filter forecasting method based on spatial state data and support vector machine forecasting method based on space-time data. On the one hand, the forecasting process is complicated, and the parameters of the model need to be set several times. On the other hand, the influence of events on the accuracy of traffic flow prediction is not considered. In view of the above problems, this paper starts with the space-time characteristics of road traffic flow under emergency, and analyzes the spatio-temporal correlation of detector traffic flow data on adjacent road sections in the event state. In this paper, a stochastic forest forecasting method considering event factors is proposed, and the validity of this method is verified by actual data. The research results of this paper are summarized as follows: (1) the traffic event acquisition method based on 122 alarm information is proposed, and the location information database of Beijing toponymic address is established. On this basis, the toponymic names of the corresponding layers and regions are extracted as the gazetteers by the way of "stratified and partitioned" toponymic data organization. At the same time, based on the spatial location information of traffic events, the traffic flow data collected by the detector are identified to know the specific status of each data. (2) based on the acquired traffic events, The collected traffic flow data are divided into traffic flow data with events and traffic flow data without events. It is found that the data of adjacent detectors have strong temporal and spatial correlation. (3) based on the temporal and spatial correlation of traffic flow and traffic events obtained, considering the factors of emergency itself, the classified data are classified. Time series data and spatial state data are used as input of stochastic forest, ARIMA and Kalman filter algorithm respectively to predict traffic flow in emergencies. (4) to improve prediction accuracy, A prediction method for spatio-temporal fusion of data is proposed. The method mainly uses the least square method to fuse the prediction results obtained by the time series data prediction method and the spatial sequence prediction method, and finally obtains a new set of prediction results. At the same time, three indexes of error evaluation are introduced: absolute value of error percentage, mean value of (MAP), absolute value of error, (MA) of absolute value of error, and (MS), of square mean of error, to evaluate the algorithm.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U491.14

【相似文献】