【作者】 池红波；

【导师】 关宏志；

【作者基本信息】 北京工业大学 ， 交通运输规划与管理， 2004， 硕士

【摘要】 短期交通预测是交通控制、车辆导航等领域需要解决的重要问题之一。研究交通预测的理论和方法，从而实现比较准确地预测未来15分钟或更长一段时间内道路上的交通状况，对于缓解城市交通堵塞、避免社会资源的浪费有着重要的意义和应用价值。 城市快速路作为城市路网中的主要经脉，承担了城市中大部分的长距离交通，快速路的交通状况在相当程度上影响着城市的出行质量。在智能交通系统的大背景下，如何利用丰富的交通检测数据，对未来的交通流状态，特别是在城市中处于重要地位的快速路的交通状态进行预测，对于提高交通运输效率，特别是物流企业的效率具有重要的理论意义和实用价值。 时间序列预测方法是在预测领域最为成熟的方法，在连续不间断的数据情况下，时间序列方法有比较高的精度。本文基于路段上的实时的观测交通流数据，利用时间序列法，结合信号处理技术和数据聚类分析理论，对城市快速路段上未来24小时单点的交通流速度预测算法进行了研究。具体包括以下工作： 对快速路段上的交通数据进行了预处理。具体包括应用交通流阈值法、交通流机理法对交通数据进行了奇异点的剔除，应用数值差分法对丢失数据进行了估计；应用了小波分析理论对数据序列进行了滤噪处理，在尽可能保证原始数据的特征的情况下，有效地过滤掉噪声数据的干扰。 利用数据聚类理论和方法对各天的路段上的交通流速度进行了聚类分析，验证了速度数据的周相似的性质，总结出了速度数据的分类表。 利用了时间序列建模的相关理论和方法对交通速度数据进行了预测建模，根据聚类分析法，分析了对历史样本的选择的方法，最后通过对结果的误差进行了分析，证明了该种选样方法的有效性。更多还原

【Abstract】 The short-time traffic forecasting is a significant problem in the field of traffic controlling, vehicle guidance and so on. It is important for us to study the method and theory on forecasting the traffic condition in the future 15 minutes or more. Well traffic prediction is valuable for alleviating traffic jam in the city and avoiding the social resource wasting.The freeway in is the artery in the urban road network, which take on most of the long-distance transportation. To a large extant the traffic condition in the freeway influences the quality of the trip in the city. On the back ground of the ITS (Intelligent Transportation System) how to make full use of the traffic data collected from the RTMS to forecast the traffic condition in the urban road especially the urban freeway, is meaningful and valuable to improve the efficiency of transportation, in particular for the logistics companies.The time-series method is a best way in the field of prediction. Under the condition of continuous data the time-series method can attain higher precision. Based on the continuous traffic data collected in the expressway of Beijing city, this thesis studies the traffic prediction method in the future 24 hours with the time-series method, contemporary signal filtering theory and data clustering theory. Concretely this studying includes such works as following:Applying the method of traffic threshold and the principle of traffic flow to eliminate the wild point in the traffic data and estimating the missing data with numeral analysis.Using the wavelet transform to denoise the data series.Employing data clustering theory to analyze the traffic flow velocity data in every day, and summarizing the regularity about the traffic speed in different days.Applying the time-series model-ARIMA this thesis forecasts the traffic speed in the future 24 hours, and analyzes the error of the result.Based on the theory of clustering, this thesis puts forward the method how to choose the proper history data to model, and analyzes the superiority of this methodover the traditional method choosing all history data.更多还原