A bi-partitioning approach to congestion pattern recognition in a congested monocentric city

Abstract

This paper proposes a bi-partitioning approach using link density data to identify an optimal restricted zone (RZ) in a congested monocentric city to assist with the implementation of area-wide congestion management strategies. A composite similarity measure is developed for each link in the network as a weighted average of a density similarity measure (ensuring homogeneity) and a distance similarity measure (ensuring connectivity and compactness). The resulting similarity matrix is fed into a graph clustering method termed symmetric nonnegative matrix factorization (SNMF) to bi-partition the network. To determine the optimal weight as a hyperparameter, we propose a hierarchical search algorithm (HSA) based on the concept of “knee” that is used to find the most significant solution from the Pareto front. The proposed approach is demonstrated on the Melbourne network using a simulation-based dynamic traffic assignment model. Results show that the methodology (i) can effectively capture the spatial variations of the congestion pattern in the network; (ii) is robust to moderate parameter changes; (iii) can be extended to the time-dependent case and hence inform when to activate the area control; and (iv) can perform relatively well in the presence of missing data. When varying the distance threshold as a design parameter, we can observe how the optimal RZ evolves in space, a feature that is critical to devising a double- or multi-layered RZ for hierarchical control purposes.