Unravelling System Optimum Structure by trajectory data analysis

Abstract

This work investigates network-related trajectory features to unravel trips that the most contribute to the system under-performance. When such trips are identified, features analysis also permits to identify the best alternatives in terms of routes to make the system to its optimum. First, data mining is carried out on trajectories obtained from reference dynamic traffic assignment (DTA) simulations in a real-world network, based on UserEquilibrium (UE) and System-Optimum (SO). This helps us (i) to target the trajectories to be changed, and (ii) to identify their main features (trip lengths, experienced travel time, path marginal costs, and network-related features such as betweenness centrality and traffic light parameters, etc.). Similarity analysis based on Longest Common Subsequence, Principle Component Analysis are the main methods that are performed to carry out descriptive analysis of trajectories. Supported Vector Machine is then used to determinate the features with regards to their contribution to better network performance.