Prediction of Urban Taxi Travel Demand by Using Hybrid Dynamic Graph Convolutional Network Model

Area of Science:

• Intelligent Transportation Systems
• Urban Mobility Analytics
• Machine Learning for Transportation

Background:

• Urban travel demand prediction is complex due to spatial-temporal dependencies and dynamic patterns.
• Existing methods often overlook dynamic demand variations and uneven distribution.
• Accurate forecasting is crucial for efficient urban transportation management.

Purpose of the Study:

• To propose a Hybrid Dynamic Graph Convolutional Network (HDGCN) for enhanced urban travel demand forecasting.
• To improve prediction accuracy by capturing dynamic spatial-temporal characteristics.
• To address limitations of existing methods in handling demand diversity and uneven distribution.

Main Methods:

• Developed a traffic demand forecasting framework using HDGCN.
• Generated dynamic graph sequences based on travel demand dynamics and time.
• Integrated dynamic and standard graph convolution modules for spatial feature extraction.
• Fused spatial features with Gated Recurrent Unit (GRU) for temporal feature learning.

Main Results:

• HDGCN demonstrated efficient and accurate prediction of urban taxi travel demand.
• The model achieved stable and effective predictions compared to state-of-the-art baselines.
• Verified using taxi data from Manhattan, New York City.

Conclusions:

• The HDGCN model effectively captures complex urban travel demand characteristics.
• It offers a significant improvement over existing methods for taxi demand forecasting.
• The framework is applicable to real-time, accurate prediction for various urban transportation systems.