Traffic Speed Prediction: An Attention-Based Method.

Area of Science:

• Intelligent Transportation Systems
• Deep Learning
• Traffic Engineering

Background:

• Short-term traffic speed prediction is crucial for intelligent transportation systems (ITSs).
• Current deep learning models often neglect spatial and environmental factors, limiting prediction accuracy.
• Understanding correlations between target and surrounding roads is vital for effective traffic management.

Purpose of the Study:

• To propose a novel traffic speed prediction approach, TCHA, integrating temporal clustering and hierarchical attention.
• To enhance prediction accuracy by accounting for diverse traffic environments and road network correlations.
• To address limitations in existing deep learning methods for traffic speed forecasting.

Main Methods:

• Applied temporal clustering to segment traffic data based on environmental similarity.
• Utilized a hierarchical attention mechanism with an encoder for spatial feature importance and a decoder for temporal feature importance.
• Evaluated the TCHA model using real-world traffic data from Hangzhou.

Main Results:

• Temporal clustering effectively distinguished traffic environments, leading to improved data distribution.
• The hierarchical attention mechanism successfully extracted relevant spatial and temporal features.
• The TCHA model demonstrated superior performance compared to state-of-the-art methods in traffic speed prediction.

Conclusions:

• The proposed TCHA method offers a significant advancement in short-term traffic speed prediction.
• Integrating temporal clustering and hierarchical attention effectively captures complex traffic dynamics.
• This approach provides a more robust and accurate solution for intelligent transportation systems.