【作者】 胡佩锋；

【导师】 袁振洲；

【作者基本信息】 北京交通大学 ， 交通运输规划与管理， 2007， 硕士

【摘要】 交通流量短时预测研究在国际上一直很活跃。在过去的几十年里,交通工程学者在这一研究领域做了大量的工作,取得了显著的成果。在此基础上,本文主要对交通流量短时预测做了进一步地研究:(1)对交通工程的研究现状进行综述,由此阐明短时交通流量预测在城市交通控制系统、交通拥挤状态自动识别和智能交通系统(ITS)的子系统交通诱导系统中的重要地位。(2)针对城市路网流量的时间和空间分布,分析了交通流量的主要特点。(3)重点介绍了现有一些交通流量预测方法,分析了各种预测方法的特点和不足。这些方法主要包括:多元线性回归方法、指数平滑方法、神经网络预测方法、时间序列预测方法和多模型融合预测法。(4)在现有预测模型的基础上,结合交通流量的特性,应用实测数据建模分析。并对时间序列模型预测交通流量的方法进行改进,提出了一种基于历史数据和预测当天实测数据的新型时间序列预测模型。该方法比当前通过一阶差分处理交通流量数据的时间序列预测方法包含信息量大,预测精度高。另外,分别采用等步长多模型融合预测方法和变步长多模型融合预测方法进行预测。最后,结合实例,对本文提出的模型和现有模型的预测结果进行对比分析。在此基础上,指出各种模型的优点、不足和需要进一步研究的问题,并对交通流量短时预测的发展进行了展望。更多还原

【Abstract】 Study on short time traffic flow (STTF) prediction is always popular in the world during the past few years.Traffic engineers had done a lot of works in this field, and obtained lots of remarkable fruits.Then futher study on STTF prediction is done in this paper.(1) We review the history and status on traffic engineering study, then, the importance of STTF prediction is introduced in Urban Traffic Control System, Automatic Congestion Identification (ACI) and Traffic Guidance Systems (TGS)–a sub-system of the Intelligent Transportation System (ITS).(2) Based on traffic volume time and space distribution on urban traffic network, main characteristics of traffic volume are introduced and analyzed.(3) In this thesis, we briefly introduce some STTF prediction methods used popular in recent years, point out their merits and shortages. The methods concluding: multiple linear regression, exponential smoothing, artificial neural networks (ANN), time series analysis and multi-model fusion algorithm (MMFA).(4) Based on these methods, traffic volume properties, and traffic volume data from traffic detectors, we build up some models and make a few analysis.An improved time series model is provided in this thesis, the model is based on both current and history data. Comparing to the traditional time series model which is just using current data (dealing with by the first difference firstly), it includes more information and has less error. Then equal step length MMFA and dynamic step length MMFA are also used to predict.At last, the results of models current using and models from this paper are compared and analysised. Then merits and shortages of these models are pointed out, questions need to be sovled are also presented, development of study on STTF prediction in the future is brief introduced.更多还原