Calibration of traffic flow fundamental diagrams for network simulation applications: A two-stage clustering approach

Abstract

This paper aims to propose a two-stage clustering approach for calibration of traffic flow fundamental diagrams for dynamic traffic assignment (DTA) simulations. Unlike previous research efforts focusing on supervised grouping strategies that are largely dependent on roadway physical attributes, a data-driven perspective is explored using big traffic data. The two-regime modified Greenshields traffic flow model is used to fit the historical observations on a daily basis using the non-linear least squares method. A two-stage clustering approach is proposed based on the calibrated models where the first stage aims to capture day-to-day variations in traffic flow fundamental diagrams while the second stage aims to aggregate links with similar traffic flow characteristics. The standard k-means algorithm is applied in the first stage and a modified hierarchical clustering based on the Frechet distance is proposed in the second stage. The calibrated and clustered results highlight the feasibility and the effectiveness of the proposed approach.