Quasi-stationary states in temporal correlations for traffic systems: Cologne orbital motorway as an example

TL;DR

This paper identifies five quasi-stationary traffic states in Cologne's orbital motorway using correlation matrix clustering, revealing non-stationary and non-Markovian features of traffic flow.

Contribution

It introduces a novel method of analyzing traffic states through clustering of reduced-rank correlation matrices, highlighting non-stationary traffic patterns.

Findings

Five distinct traffic states identified, including holiday and workday states.

Workday and mixed states show strong correlated time groups.

Traffic states mapped onto velocity correlation matrices revealing free and congested conditions.

Abstract

Traffic systems are complex systems that exhibit non-stationary characteristics. Therefore, the identification of temporary traffic states is significant. In contrast to the usual correlations of time series, here we study those of position series, revealing structures in time, i.e. the rich non-Markovian features of traffic. Considering the traffic system of the Cologne orbital motorway as a whole, we identify five quasi-stationary states by clustering reduced rank correlation matrices of flows using the $k$-means method. The five quasi-stationary states with nontrivial features include one holiday state, three workday states and one mixed state of holidays and workdays. In particular, the workday states and the mixed state exhibit strongly correlated time groups shown as diagonal blocks in the correlation matrices. We map the five states onto reduced-rank correlation matrices of velocities and onto traffic states where free or congested states are revealed in both space and time. Our study opens a new perspective for studying traffic systems. This contribution is meant to provide a proof of concept and a basis for further study.