An urban commuters' OD hybrid prediction method based on big GPS data

Area of Science:

• Transportation Science
• Data Science
• Urban Planning

Background:

• Rapid advancements in mobile internet and GPS technology generate vast amounts of trajectory data.
• Urban commuters' travel information within trajectory data is valuable for traffic studies and urban mobility analysis.

Purpose of the Study:

• To develop a novel hybrid prediction method for urban commuters' origin-destination (OD) volume.
• To simultaneously consider temporal and spatial dependencies within big GPS trajectory data.

Main Methods:

• Regional division using a grid map to construct OD pairs and establish network topology.
• Application of a graph convolutional network (GCN) for spatial dependencies and a long short-term memory (LSTM) network for temporal dependencies.
• Integration of an attention mechanism to learn input data weights.

Main Results:

• The proposed hybrid OD prediction method demonstrated significant performance.
• Numerical experiments using GPS data from Chengdu, China, showed reasonable prediction accuracy.
• Comparative analysis validated the effectiveness of the hybrid approach.

Conclusions:

• The developed hybrid method offers a robust approach for predicting urban commuter OD patterns.
• The integration of deep learning techniques (GCN, LSTM) and attention mechanisms enhances prediction accuracy.
• This research contributes to better understanding and managing urban traffic dynamics using big data.