【作者】 刘倩；

【导师】 张冬泉；

【作者基本信息】 北京交通大学 ， 安全技术及工程， 2009， 硕士

【摘要】 现有城市轨道交通客流诱导系统采用基于应急预案的广播加人工疏导方式解决了突发事件下的客流疏散问题,对于乘客日常出行路径诱导的研究则存在空白。随着城市轨道交通路网规模的扩大,缺少乘客日常出行路径诱导将导致大量乘客集中在OD间里程或时间最短路径的情况,造成车厢拥挤和换乘拥堵等问题。本文参考城市道路交通诱导系统的相关理论,结合轨道交通的特点,以北京城市轨道交通系统为蓝本,进行了相关的理论研究及实践,主要工作及结论如下:1.对交通诱导系统基础理论,即最优路径选择模型进行研究,建立了包含确定和非确定因素的综合出行阻抗函数,并使用Java优先级队列完成Dijkstra算法的O(NlogN)实现。2.通过对交通分配理论的研究,确定以最短路分配为理论指导,使用Java实现北京城市轨道交通系统以5分钟为粒度的OD客流在断面上的分配。3.通过对北京城市轨道交通系统大量进出站以及断面历史客流样本进行分析,得出使用最小二乘拟合能够描述工作日客流分布规律的结论。4.指出对于BP神经网络,选择导函数取值范围大的传递函数能带来更快的训练收敛速度;对比12种常见BP网络训练算法并总结其各自在训练收敛速度以及内存消耗方面的优劣,特别强调了一种改进的共轭梯度算法——SCG算法的广泛适用性;指出使用训练提前终止(Early Stopping)技术对于保证BP网络训练结果推广能力的作用。5.以北京城市轨道交通系统2号线西直门站工作日进出站客流为例,分别研究基于日期和时段时间序列的BP神经网络预测模型。通过大量试验确定适用于客流预测的BP网络拓扑结构,并采用Tan-Sigmoid型传递函数、SCG算法以及Early Stopping技术建立收敛速度快、内存需求低、泛化能力强的BP神经网络。在一定误差范围内揭示了历史客流与短时未来客流之间的关系。6.应用新型SSH框架,即以Hibernate作为持久层实现,Struts作为表示层实现,结合Spring作为业务逻辑层实现与框架整合的一套高内聚、低耦合、易维护的轻量级J2EE Web开发框架,结合Flash矢量动画技术,实现了以图形交互方式综合展现客流信息与诱导方案的高质量B/S模式北京城市轨道交通客流诱导软件系统。更多还原

【Abstract】 Existing passenger flow guidance system of urban rail transit is based on emergency preparedness. It is about using broadcasting and artificial grooming to solve unexpected problems of passenger flow. But, there are gaps in the field of day-to-day travel path guidance for urban rail transit. With the expansion of the transport network, a large number of passengers will get together in the path having best distance or time between OD without guidance. A big problem of congestion will appear in that case. So, both related road traffic theory and the characteristics of rail transit are taken into consideration. Main work of this paper is carrying out research and realization with urban rail transit system in Beijing as background. The contents of this paper can be summarized as follows:1. On the basis of research for optimum route search model ( basic theory of traffic guidance system), comprehensive travel impedance function including both deterministic and non-deterministic factors and a O(NlogN) Dijkstra algorithm implementation with Java PriorityQueue is given.2. After traffic assignment theory study, minimum path assignment model is choosed to achieve calculation of 5-minute interval section passenger flow. It is realized by assigning OD passenger flow to section with Java program.3. A large numbers of historical station and section passenger flow samples are analysed to determin the characteristics of passenger flow distribution. The result turns out that fitting passenger flow distribution by least square method is a reasonable scheme.4. Some conclusions about BP neutral networks are summarized: Transfer function whose derivative function has larger value range will perform better convergence speed in training. A scaled conjugate gradient algorithm (SCG) is emphasized after comparison of common BP training algorithms about training convergence speed and memory consuming. Early Stopping is a useful technology to improving generalization.5. Take passenger flow into and out of a station of Beijing urban rail transit system called XiZhiMen for example, this paper discusses BP neutral network model for date and time prediction from aspect of times series. Trial-and-error method is still the best way to determine the topological structure of BP net. A BP neutral network with high convergence speed, low memory consumption and good generalization is established by Tan-Sigmoid transfer function, SCG algorithm and Early Stopping technology. The mapping relationship between passenger flow of history and that of short-term future within a certain margin of error can be revealed by this BP neutral network.6. A new integrated framework called SSH is used in this paper to achieve a kind of J2EE Web Development with high cohesion, low coupling and easy maintenance. SSH framework is made of Hibernate (for persistence layer), Struts (for presentation layer) and Spring (for business logic layer and framework integration). Finally, a high quality passenger flow guidance software system of Beijing urban rail transit with B/S model is made by this SSH framework combine with Flash technology for vectorgraph and animation. Both passenger flow information and guidance result can be displayed by graphic interaction.更多还原