Traffic Flow Assignment on Expressways: A GCN‐TODIM Framework Leveraging Large‐Scale Trajectory Data

Abstract

As expressway networks expand and road system complexity increases, the demand for accurate traffic assignment results continues to grow. However, traditional models struggle to capture the behavioral inertia of travelers, limiting assignment accuracy. This study proposes a novel traffic assignment model integrating graph convolutional networks (GCNs) with the multicriteria decision‐making method TODIM. Vehicle trajectory data from expressway toll gantries enable GCN to extract route choice inertia, whereas TODIM incorporates travel time, distance, and cost to simulate dynamic path selection. An iterative optimization process achieves traffic flow equilibrium. The model was validated on expressway networks in Jinan, Taian, and Weifang, Shandong Provinces. Results show that during peak hours, the GCN‐TODIM (G‐T) model outperforms stochastic user equilibrium (SUE), bounded rational user equilibrium (BR‐UE), and mean excess travel time (METT), reducing mean absolute error by 36.1%–52.4% and root‐mean‐square error by 27.3%–43.5% and improving R 2 by 13.9%–16.6%. Similar improvements were seen off‐peak, confirming strong accuracy and robustness, especially in networks with many segments and ramps. Overall, the G‐T model offers an efficient, practical tool for traffic assignment, supporting more scientific highway planning and management strategies.