Hybrid Deep Learning Approach for Traffic Speed Prediction.

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Data Science

Background:

• Accurate traffic speed prediction is crucial for traffic management and route planning.
• Existing models struggle to simultaneously capture complex spatial and temporal correlations in traffic data.
• This limitation leads to suboptimal traffic speed prediction performance.

Purpose of the Study:

• To propose a novel hybrid deep learning approach, HDL4TSP, for accurate traffic speed prediction.
• To effectively model both spatial and temporal dependencies in urban traffic data.
• To enhance the performance of traffic speed prediction systems.

Main Methods:

• A hybrid deep learning architecture (HDL4TSP) comprising input, spatial, temporal, fusion, and output layers.
• Graph convolutional networks (GCNs) in the spatial layer to capture near and distant spatial dependencies.
• Convolutional Long Short-Term Memory (ConvLSTM) networks in the temporal layer to model daily and weekly periodicities.

Main Results:

• The proposed HDL4TSP model successfully integrates spatial and temporal features.
• Extensive experiments demonstrated superior performance compared to four baseline methods.
• The model achieved significant improvements in traffic speed prediction accuracy on two real-world datasets.

Conclusions:

• HDL4TSP offers a robust solution for traffic speed prediction by effectively handling spatial and temporal correlations.
• The hybrid deep learning approach provides a significant advancement over existing methods.
• This model has strong potential for practical applications in intelligent transportation systems.