Dynamic partitioning of urban road networks based on their topological and operational characteristics

Modelling, observation and control of realistic urban road networks of significant size may be subject to the `curse' of dimensionality. On the other hand, the `nature' of the physical system introduces the necessity of its unified treatment, for capturing the interactions and the spatial connectivity of the complete system. In order to facilitate the handling of such complex traffic systems, an approach could be the optimal partitioning of the complete network in sub-regions/networks based on one-or more-criteria. The resulted sub-networks then can be processed more easily (e.g. utilizing tactics of parallel processing), increasing computational performance. Several methods are available and can be applied for partitioning graph/networks, though road networks exhibit particularities that pose constraints in applying standard mesh/graph partitioning technics. One of the most important features stands for the fact that road networks are directed networks with many bidirectional arcs of dynamic operational characteristics, which should be partitioned in compact closed sub-regions. In the current paper the results of an investigation on dynamic network partitioning are provided and discussed in detail. A data-based clustering method has been initially tested, namely k-means, accounting for network's topology, structure and operational characteristics, exhibiting the strengths and restrictions of its application in realistic and dynamic settings. However, the fact that k-means is not a method dedicated in providing road network's partitions, the use of a graph partitioning model, namely METIS. The data used here have been collected from the real-time surveillance system operated in Nicosia, Cyprus, that provides detailed traffic observations (flow and speed), while the comparative results of the two approaches are also provided. It has been observed that METIS partitioning could be used for network partitioning purposes, in the various alternative datasets used for performing a dynamic network partitioning `exercise'. The results showed the effect of traffic dynamics on partitioning realistic urban systems.