Recognizing the Traffic State of Urban Road Networks: A Resilience-Based Data-Driven Approach

Accurate and timely traffic network state recognition is crucial in supporting intelligent transportation system (ITS) urban traffic control and guidance. Despite their significance, existing methods for traffic state recognition often fall short of practical demands owing to the dynamic nature and unpredictability of traffic flows and the high costs associated with sample processing. This paper introduces a novel resilience-based approach for classifying and identifying link-level traffic states in urban road networks by focusing on these challenges. This approach has two phases: 1) the classification phase introduces a new operational resilience index and uses a hybrid K-means++-fuzzy c-means (FCM) clustering method for traffic state labeling; and 2) the identification phase employs real-time automatic vehicle identification (AVI) data and a transformer-based model to determine current traffic conditions. A case study conducted by Shaoxing validated the effectiveness of this approach. The results show that the objective function value of the hybrid clustering method is 0.168, with a classification performance metric Xie-Beni (XB) index of 0.137 and a Davies-Bouldin index (DBI) of 12.39, indicating high-quality clustering. A comparative analysis with support vector machines, convolutional neural networks, and long short-term memory (LSTM) models revealed the superior identification performance of the transformer-based model, which achieved 93.35% accuracy (increases of 21.44%, 13.01%, and 5.89%, respectively). The proposed method offers a practical reference for real-time traffic condition monitoring from a resilience perspective in traffic management systems.