Data-Driven Traffic Assignment Through Density-Based Road-Specific Congestion Function Estimation

Abstract

The ability to build accurate traffic assignment models on large-scale major road networks is essential for effective infrastructure planning. Static traffic assignment models often utilize standard formulations of congestion functions which suffer from various inaccuracies. Conversely, newer approaches in the literature rely on inverse optimisation to provide enhanced accuracy but incur significantly heavy computational costs. The work in this article develops density-based congestion function fitting in order to compute traffic assignment patterns. Computational efficiency makes the method amenable to be used on real-world networks at national scale. The methodology is applied on the motorway network connecting the primary metropolitan areas in England using Motorway Incident Detection and Automatic Signalling system data. The results demonstrate that the use of density-based congestion functions provides significant improvement in terms of computational runtime in the order of 11,000 times (22 secs vs 68 hours). Correspondingly, prediction error from this method (3.9 to 6.9% for time prediction and 10.4 to 10.7% for flow prediction) slightly outperforms the state-of-the-art Inv-Opt method (5.3 to 8.8% for time prediction and 10.5 to 11% for flow prediction). The increased accuracy provides greater confidence in modelling results for applications such as cost-benefit analysis and price of anarchy calculations.