Understanding Private Car Aggregation Effect via Spatio-Temporal Analysis of Trajectory Data.

Area of Science:

• Transportation Science
• Artificial Intelligence
• Urban Planning

Background:

• Understanding private car aggregation is crucial for intelligent transportation and urban planning.
• Weekend private car mobility presents challenges due to random patterns and inefficient spatiotemporal feature representation.

Purpose of the Study:

• To propose a deep learning framework, STANet-NALU, for understanding the dynamic aggregation effect of private cars on weekends.
• To enhance the representation of spatiotemporal features for improved prediction accuracy.

Main Methods:

• Developed a spatiotemporal attention network (STANet) integrated with a neural algorithm logic unit (NALU).
• Designed an improved kernel density estimator (KDE) with a log-cosh loss function for robust spatial distribution calculation.
• Utilized private car stay time as a temporal feature to capture nonlinear temporal correlations.
• Implemented a spatiotemporal attention module and a gate control unit for adaptive feature fusion.

Main Results:

• The proposed STANet-NALU framework demonstrated superior performance compared to existing methods.
• Achieved significant improvements in metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Kullback-Leibler divergence (KL), and R2.
• The model exhibited strong numerical extrapolation capabilities for predicting weekend private car aggregation.

Conclusions:

• STANet-NALU effectively models the dynamic aggregation effect of private cars on weekends.
• The framework offers a robust and accurate approach for intelligent transportation management and urban planning.
• This research provides valuable insights into weekend mobility patterns for enhanced city infrastructure development.