Multitask Learning with Graph Neural Network for Travel Time Estimation.

Area of Science:

• Artificial Intelligence
• Transportation Science
• Data Science

Background:

• Accurate travel time estimation (TTE) is crucial for ride-hailing and navigation.
• TTE is influenced by complex static and dynamic spatial-temporal factors.
• Existing methods struggle to capture the intricate relationships between these factors.

Purpose of the Study:

• To develop an advanced framework for precise travel time estimation.
• To address the challenges in representing and modeling spatial-temporal dependencies in TTE.
• To improve the accuracy of TTE for individual road segments and entire paths.

Main Methods:

• A novel framework integrating graph convolutional neural networks (GCN) and recurrent neural networks (RNN) for segment-level TTE.
• A graph attention network (GAT) to model the influence of adjacent road segments on TTE.
• A multitask learning model for simultaneous estimation of path and segment travel times.

Main Results:

• The proposed method achieved a percentage estimation error of 13.91% on real-world taxi trajectory datasets.
• The framework demonstrated significant performance improvements over three state-of-the-art methods.
• Effective capture of spatial-temporal dependencies and inter-segment relationships was validated.

Conclusions:

• The developed framework offers a robust solution for accurate travel time estimation.
• The integration of GCN, RNN, and GAT effectively models complex spatial-temporal dynamics.
• This approach holds significant potential for enhancing ride-hailing dispatch and navigation systems.