Modeling and Analysis of Congested Systems Consisting of Intelligent Customers

在这项研究中，我们研究了拥挤现象，在这种现象中，客户的行为是智能的。到目前为止，人们主要是利用排队模型和汽车交通模型来研究和分析拥堵现象，例如在电信系统、计算机系统生产线和汽车交通中。在这些现象中，顾客的行为在一定程度上受到物理约束，因此我们可以假设顾客在模型中遵守简单的规则，但对于其他类型的拥堵现象，我们可能不得不在模型中考虑更复杂的顾客行为或判断，因为顾客在这些现象中有更多的自由。在这里，我们考虑了以下现象：a)行人在十字路口的行走)乘客在上、下电车时的动作)用户在互联网终端的行为)用户在手机上的行为a)每个行人预测未来的碰撞，并为…选择最佳方向和最佳速度更多的步行而不会有任何碰撞。利用该模型，通过仿真实验，推导出行人密度与平均车速之间的关系。这种关系与从真实的行人数据中观察到的关系非常相似。这表明我们的模型捕捉到了行人行为的主要方面。在b)中，我们开发了一个模型，其中每个乘客试图最小化他/她的潜在函数，该函数反映了其他乘客的距离和视线。此外，该模型还纳入了与上下车有关的激励措施。利用该模型，我们通过仿真得出了列车拥堵程度与一定数量乘客下车所需时间之间的关系。结果表明，当有一半乘客下车时，时间比所有乘客下车时的时间要长。这种现象已经是已知的，但这可能是第一个对这种现象进行量化评估的结果。在c)中，我们观察了通过东京工业大学局域网大门的流量数据，并建立了流量模型。然而，由于某些原因，该模型不能充分反映用户的行为。这是一个有待进一步研究的主题。d)由于无法获得合适的数据，因此很难捕获用户的行为。较少

In this research, we study congested phenomena in which customers behave with intelligence. So far congested phenomena, for example in telecommunication systems, computer systems production lines and car traffics, have been studied and analyzed mostly using models such as queuing models and car traffic models. In these phenomena, customers' behaviors are somewhat restricted by physical constraints, and so we may assume customers obey simple rules in the models.However for congested phenomena of other types, we might have to take more complicated customers' behaviors or judgements into account in our model because customers have more freedom in these phenomena. Here we consider the following phenomena :a) pedestrians' walks on a crossingb) passengers' movements at getting on and off a trainc) users' behavior at internet terminald) users' behavior at mobile phonesIn a), we develop a model in which each pedestrian anticipates future collisions and chooses best direction and best speed for … More walking without any collisions. Using this model, we derive a relation between pedestrian density and mean speed through simulation experiments. The relation is very similar to the one observed from real data of pedestrians. This shows our model captures major aspects of pedestrians' behavior.In b), we develop a model in which each passenger tries to minimize his/her potential function which reflects the distances from and the sights of other passengers. Further the model incorporates incentives concerning to getting on/off. Using this model, we derive a relation through simulation "between the congestion of a train and the time required for some number of passengers to get off. The result shows that when the half of passengers get off the time is longer than the case where all passengers get off. This phenomena has been already known but probably this is the first result which evaluate the phenomena quantitavely.In c), we observe traffic data passing through the gate of LAN of Tokyo Institute of Technology and develop a model for the traffic. However, the model cannot reflect users' behavior sufficiently for some reason. This is a theme for future study.In d), it is very difficult to capture users' behavior since we cannot get suitable data. Less