Transformer model-based multi-scale fine-grained identification and classification of regional traffic states

Area of Science:

• Intelligent Transportation Systems
• Machine Learning in Transportation
• Traffic Flow Theory

Background:

• Conventional traffic state estimation methods lack precision, especially in capturing transitions after congestion reduction strategies.
• Existing methods often use limited categories (4-6), failing to represent nuanced traffic dynamics.
• Effective transition from congested to free-flowing states remains a challenge for current strategies.

Purpose of the Study:

• To develop a novel Transformer-based approach for precise traffic state identification and classification.
• To enhance the granularity of traffic state representation by introducing a 16-level classification system.
• To improve the accuracy of traffic state estimation, particularly during and after congestion mitigation.

Main Methods:

• A Transformer model was designed to extract salient features from traffic data.
• The k-means clustering algorithm was applied to group similar traffic states based on extracted features.
• Non-dominated sorting was used to rank the clusters, establishing 16 distinct congestion levels (Level 1 to Level 16).

Main Results:

• The Transformer-based approach demonstrated significant advancements in traffic state estimation.
• Experimental results on a large-scale simulated dataset showed marked improvements in clustering quality.
• The proposed model exhibited superior generalization capabilities compared to baseline methods.

Conclusions:

• The Transformer model offers a more precise and granular method for traffic state estimation.
• The 16-level classification framework effectively captures nuanced traffic conditions.
• This approach provides a robust solution for intelligent transportation systems seeking to optimize traffic flow and management.