【作者】 杨军；

【导师】 侯忠生；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2014， 博士

【摘要】 在地铁客流组织与优化研究领域,客流短期预测和行人运动模拟是其关键技术。虽然不少研究学者已开展了相关工作,但如何适应轨道交通业务发展需要提高客流短期预测效果,对重大事件引发的大客流进行实时预测,以及由于大客流和重大事件引发对地铁站内不同场景进行乘客疏散模拟等问题亟待解决。本文的内容要点如下：(1)提出了一种基于小波分析的支持向量机客流预测算法,首先对原始客流时间序列数据进行小波分解,然后利用最小二乘支持向量对分解得到的低频和高频信息进行学习与预测,最后用小波合成重构低频预测信号与高频预测信号,得到预测客流时间序列数据。实验采用北京市轨道交通客流数据和标准评价方法,结果表明该算法具有较好的预测效果,且优于两种常见的客流预测算法。(2)构建了一种基于灰色马尔科夫的大客流实时预测算法,利用灰色预测算法对客流数据建立灰色模型,然后建立马尔科夫修正模型,最后利用预测误差对灰色预测结果进行修正得到大客流预测值。实验针对多种类型的大型活动和重大节假日进行大客流实时预测,结果表明该模型对真实的重大事件大客流预测效果较好。(3)建立了一种面向多向行人疏散流的向量地场模型,模拟地铁站内行人多向行走和疏散过程。鉴于行人对于不同方向的敏感程度是不同的,该模型着重考虑方向的影响,在向量地场模型中更细致地表现在不同方向行人之间的相互影响。仿真实验表明,该模型能较好地模拟和再现地铁站内不同场景下多向行人之间的复杂相互作用和自组织现象。(4)由于正常情况下行人通行时表现出一种相互排斥作用,因此本文在传统地场模型的基础上,用排斥作用代替了适宜模拟紧急疏散情况的跟随作用,建立了一个引入排斥动态场的行人疏散地场模型。仿真结果表明,适当的排斥力作用可提高模拟场景的拥堵临界密度。更多还原

【Abstract】 ABSTRACT:In the research field of Metro passenger flow organization and optimization, short-time passenger flow prediction and pedestrian movement simulation are significant technologies. Although a lot of researchers have done some related works, several difficulties, such as how to increase the passenger forecasting performance with the metro transportation business, predict the real-time passenger flow for important events effectively, and simulate the pedestrian evacuations in different metro scenes for some special cases, need to be improved. The main works in this dissertation include:1. A wavelet based support vector machine short-time passenger prediction algorithm is presented. This method firstly decomposes the original passenger flow time series into low-frequency and high-frequency information by wavelet, and then the least squares support vector machine method is applied to learn from the decomposed information and then to forecast, and finally the predicted low-frequency and high-frequency time series are reconstructed to obtain the predicted passenger flow data. The experiments are based on the passenger flow data from Beijing Metro system and standard evaluation methods, and the results show that the proposed algorithm has better prediction effectiveness than the two existing forecasting methods.2. This paper introduces a grey Markov based larger passenger flow real-time prediction method, and this algorithm establishes the grey model for passenger time-series data by grey prediction method at first, and then constructs the improved Markov model, and grey forecasting results are corrected by the prediction error to acquire the predicted large-scale passenger flow finally. Based on several kinds of large-scale activities and major holiday events, the experimental results indicate that the presented model has achieved better predicting effectiveness for large-scale passenger flows in important events.3. A vector floor field model for mixed movement with multi-direction pedestrian evacuation flow is proposed. It is used to simulate the pedestrian dynamic process in subway stations. As the pedestrians’ sensitivities in different direction are diverse, this model considers the direction factor, and the interactions among pedestrians with different directions can be reproduced much better in the vector floor field model. The results show that this model could realistically and effectively simulate and reproduce the complex interactions and self-organization phenomena among multi-directional pedestrians under a variety of subway station scenarios.4. Considering the fact that the interactions among pedestrians are regarded as mutual repllsion at normal circumstance, a new floor field model for pedestrian evacuation movement with dynamic repulsion field is presented, and in this model the repulsion is used to instead the following effect widely applied in the emergency evacuation. The results show that the appropriate repulsion can improve the congestion critical density in simulation scenarios.更多还原