Spatial-Temporal Attention Mechanism and Graph Convolutional Networks for Destination Prediction

Area of Science:

• Intelligent Transportation Systems
• Urban Mobility
• Data Science

Background:

• Urban transportation destination prediction is vital for traffic planning and congestion control.
• Existing methods struggle with the dynamic and complex spatial-temporal nature of traffic data.
• Accurate modeling of road network nonlinearity and traffic flow dynamics is challenging.

Purpose of the Study:

• To propose a novel human-in-loop Spatial-Temporal Attention Mechanism with Graph Convolutional Network (STAGCN) model.
• To effectively model and capture complex spatial-temporal dependencies in urban traffic data for destination prediction.
• To improve the accuracy of urban car-hailing destination prediction.

Main Methods:

• Representing the traffic network as a graph using grid region division.
• Applying graph convolutional operations to learn spatial-temporal correlations.
• Utilizing an attention mechanism to analyze periodic features and enhance key node representations.
• Integrating spatial and temporal features as input for a Long-Short Term Memory (LSTM) network.

Main Results:

• The proposed STAGCN model demonstrates superior performance in urban car-hailing destination prediction.
• Experimental results on a large-scale urban real dataset validate the model's effectiveness.
• STAGCN achieves better prediction accuracy compared to traditional baseline models.

Conclusions:

• The STAGCN model offers a robust solution for urban destination prediction by effectively modeling spatial-temporal dependencies.
• The integration of graph convolutional networks, attention mechanisms, and LSTMs enhances prediction accuracy.
• This approach provides a significant advancement for intelligent transportation systems.