TSTA-GCN: trend spatio-temporal traffic flow prediction using adaptive graph convolution network.

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Metro passenger flow prediction faces challenges in balancing long-term and short-term demands.
• Accurately modeling spatial and temporal dependencies is crucial for effective metro transit management.

Purpose of the Study:

• To develop an advanced model for metro passenger flow prediction.
• To address the complexities of spatio-temporal dependencies in transit data.
• To improve the accuracy of both short-term and long-term passenger flow forecasts.

Main Methods:

• A Trend Spatio-Temporal Adaptive Graph Convolution Network (TSTA-GCN) model was developed.
• A trend convolutional self-attention mechanism was employed to learn temporal trends.
• Adaptive graph convolution and spatio-temporal interaction modules were utilized to capture spatial and dynamic correlations.

Main Results:

• The TSTA-GCN model demonstrated superior performance compared to state-of-the-art baseline methods.
• The model effectively predicted both short-term and long-term metro passenger flow.
• Experimental results validated the model's capability in handling spatio-temporal heterogeneity.

Conclusions:

• The TSTA-GCN model offers a robust solution for metro passenger flow prediction.
• The proposed method enhances the understanding and forecasting of urban transit dynamics.
• This research contributes to optimizing metro operations through accurate passenger flow analysis.