【作者】 牛虎；

【导师】 关伟；

【作者基本信息】 北京交通大学 ， 系统工程， 2010， 硕士

【摘要】 公交车辆到站时间是出行者最为关心的交通信息之一,提高公交车辆到站时间预测模型的精度和可靠性,可对城市公共交通的发展起到积极的推动作用。论文系统地分析了公交车辆到站时间的组成部分和影响因素,选取公交车辆在前续站点的到站时间、停靠时间和延误程度为预测模型的输入变量,设计并实现了基于车辆GPS数据的插值方法、数据库结构和数据处理算法,得到了公交车辆在每站的详细运行数据。在此基础之上,论文首先提出了基于平均行驶和停靠时间的统计模型；随后建立了BP人工神经网络预测模型,使用样本数据来训练神经网络,拟合前续站点到站时间、停靠时间和延误程度与后续到站时间之间的非线性关系,利用实时的已到站信息对后续到站时间进行预测；最后提出了改进的非参数回归模型,先对搜索数据库进行聚类分析,随后对状态向量进行主成分分析以达到降维的效果,在此基础之上选取K近邻机制和加权平均预测算法来构建模型。最后论文对北京公交16路实际采集来的3369组数据进行实证分析,以上行方向为例,详细分析了数据特性,选取相对平均误差MRE作为评价模型预测效果的指标,对三个模型进行了计算,发现非参数回归预测模型在最优K值条件下的平均预测误差最小,相对于统计模型和人工神经网络模型分别改善了55.15%和40.24%。更多还原

【Abstract】 Bus arrival time is one of the most concerned traffic information for travelers and it is a very important subject to improve the precision and reliability of the prediction model which can promote the development of city public transportation.This thesis analyzes the components of bus arrival time systematically and the bus arrival time, dwell time and stop delay at previous stops are chosen as the main input variables of the prediction model. After that this thesis designs the algorithm of data interpolation and processing in order to get the input variables of the prediction models.Based on the processed data and analysis, this thesis firstly puts forwards the statistical model based on average running time of each link and dwelling time of each stop. After that BP-Artificial Neural Network (BP-ANN) is modeled and the mass sample data are used to train the non-linear relationships between the input and output variables. With the trained network and real-time input variables of the previous stops, the bus arrival time of the coming stops can be forecasted. In the end, this thesis uses the improved non-parameter regression method by using the principal component analysis and clustering methods. Then the specified procedures and steps of the improved model are designed by using K nearest neighbor mechanism and weighted arithmetic mean method.In the end this thesis gives the case study of the 3369 group data from line 16 of Beijing Bus Company. It analyzes the data characteristics and uses the index of Mean Relative Error (MRE) to evaluate the three models. It is founded that the non-parameter regression model of the optimum K value can obtain the best average prediction result which has improved by 55.15% and 40.24% compared to statistical model and BP-ANN respectively.更多还原