Related Experiment Video
Updated: Jan 11, 2026
05:55
Modeling the Functional Network for Spatial Navigation in the Human Brain
Published on: October 13, 2023
1.5K
Deep Learning-Driven Bus Short-Term OD Demand Prediction via a Physics-Guided Adaptive Graph Spatio-Temporal
Zhichao Cao1,2, Longfei Song1, Silin Zhang1
1School of Transportation and Civil Engineering, Nantong University, Nantong 226019, China.
Sensors (Basel, Switzerland)
|November 13, 2025
Summary
This study introduces the physics-guided adaptive graph spatio-temporal attention network (PAG-STAN) for predicting bus origin-destination (OD) demand. The model achieves superior accuracy by integrating physics-guided mechanisms and attention networks.
Area of Science:
- Transportation Science
- Artificial Intelligence
- Deep Learning
Background:
- Accurate short-term bus origin-destination (OD) demand prediction is crucial for public transit operations.
- Existing models often struggle with limited data and capturing complex spatiotemporal dependencies.
- A novel deep learning approach is needed to enhance prediction accuracy and efficiency.
Purpose of the Study:
- To develop and evaluate a novel deep learning model, PAG-STAN, for short-term bus OD demand prediction.
- To improve prediction accuracy using a physics-guided loss function and attention mechanisms.
- To validate the model's performance on a small-scale dataset with a 30-minute interval.
Main Methods:
- Constructed a modified deep learning model (PAG-STAN) integrating physics-guided mechanisms, adaptive graph convolution, attention networks, and a spatiotemporal encoder-decoder.
- Employed an encoder-decoder architecture with adaptive graph convolutional LSTM (AGC-LSTM) and a multi-head attention mechanism.
- Utilized a masked physics-guided loss function incorporating boarding passenger volume and OD demand, trained with Adam optimizer and early stopping.
Main Results:
- PAG-STAN demonstrated superior prediction accuracy compared to other deep learning models.
- The model achieved significant reductions in Root Mean Square Error (RMSE) by 6.19%, Mean Absolute Error (MAE) by 6.59%, and Weighted Mean Absolute Percentage Error (WMAPE) by 8.20%.
- An improvement of 1.13% in R-squared (R²) was observed, indicating enhanced predictive power.
Conclusions:
- The proposed PAG-STAN model effectively predicts short-term bus OD demand, outperforming existing methods.
- Integrating physics-guided principles and attention mechanisms enhances the model's ability to capture complex spatiotemporal patterns.
- The study highlights the potential of advanced deep learning techniques for optimizing public transportation management.
