Room 409, Building #1, the University of Tokyo

The 8th of International BinN Research Seminar “Dynamic Behavior Analysis and Clustring in Unsteady Networks”

The 8th International BinN Research Seminar “Dynamic Behavior Analysis in Unsteady Networks” will be held on July 5th 2015. As keynote speakers, we will invite Dr. Konstantinos Ampountolas from University of Glasgow. Dr. Dr. Konstantinos Ampountolas is currently doing research on network analysis and in the seminar, keynote lectures would focus on functional distributional algorithm for clustering heterogeneous traffic networks using spatiotemporal data.  In addition, we discuss about new approaches of unsteady behavioral modeling with two researchers’ presentation.


Ashwini Venkatasubramaniama,b,c, Ludger Eversa, and Konstantinos Ampountolas*, School of Mathematics & Statistics, Urban Big Data Centre (UBDC) University of Glasgow, UK


Functional distributional clustering of traffic networks for spatio-temporal data


Clustering analysis provides a selection of a finite collection of templates that well represent, in some sense, a large collection of data. Nowadays clustering has many applications in engineering, computer science, social and life sciences, due to the availability of large volumes of data from user-generated content and emerging infrastructure-based sensors. In this talk, we present a functional distributional algorithm for clustering heterogeneous traffic networks using spatiotemporal data. The proposed algorithm seeks to identify spatially contiguous clusters in Manhattan-like grid networks and has the ability to accommodate temporal data with bi-modal characteristics. The algorithm draws on a measure of distance that utilises (cumulative distribution) functions of observations rather than functions of clusters. We describe methods to determine the optimal number of clusters within a hierarchical agglomerative clustering framework. This helps to evaluate the similarity between distinct identified clusters and “true” clusters to measure the algorithm’s performance. Results demonstrate that the proposed functional distributional clustering algorithm has a greater ability to efficiently identify clusters compared to functional only and temporal only algorithms. On-going work on dynamic clustering seeks to identify clusters that change over time.

Sachiyo Fukuyama

Department of Civil Engineering, University of Tokyo

Title: Network analysis for urban planning based on the historical development process


We propose a method of network analysis to figure out the spatial structure and characteristics of urban districts, which are assumed to be important for efficient urban planning and renovation. We use a simple index that reflect route choice behavior for analyzing road networks in the periods before behavioral surveys started. For a case study, we apply the method to the historical networks of the old city of Barcelona and find the relation between the streets of high centrality and the placement of open spaces.

Eiji Hato and Samal Dharmarathna*

Department of Civil Engineering, University of Tokyo



Unsteady travel behavior under uncertainty in densified networks


Understanding the travellers’ behavior under uncertainty is essential to minimize the congestion and maintain the service level of densified networks during unexpected events such as earthquakes or extreme weather events. During such events, drivers’ pre-trip decisions are get disturbed and it becomes quite obvious to assume that their cognition and decision-making mechanisms are more myopic as the network condition is likely to be stochastic. But still there is some space that drivers could use their spatial knowledge on the network to choose the route.

This on-going study tries to cope with both these concepts by using the generalized recursive logit (GRL) model and compare the differences, by using the probe taxi data collected in Tokyo during the period of Great East Japan Earthquake occurred on 11th March 2011 and torrential rain occurred on 23rd July 2013. Gridlock phenomena has occurred in Tokyo for the first time, after the earthquake due to the temporary shutdown of the metropolitan expressway and all railways for checking purposes. The behavior of the sequential discount rate which generalize the drivers’ decision making dynamics and represent the degree of spatial recognition of network as a parameter is compared along with other parameters such as travel time and right turn dummy within the event by using similar data collected exactly one week before and after the earthquake respectively on 04th and 18th of March 2011. During the event of torrential rain, some of the links that has under passes and depressions were inundated and the cars or taxies couldn’t move across. Hence the travellers’ use such routes under normal circumstances had to choose alternative routes. In this case also, the aforementioned parameters were estimated and compared within the event by using the similar data collected exactly one week after the event on 30th July 2013. In addition, we would like to present the comparison of parameters between the two events as well.