Vehicle Trajectory Prediction via Urban Network Modeling.

Area of Science:

• Artificial Intelligence
• Transportation Science
• Urban Planning

Background:

• Empty taxis represent significant wasted resources in urban transportation systems.
• Current trajectory prediction methods often neglect crucial spatial information, leading to inefficiencies.
• Alleviating supply-demand imbalance and traffic congestion necessitates accurate real-time taxi trajectory prediction.

Purpose of the Study:

• To develop a novel model for accurate taxi destination prediction.
• To address the limitations of existing methods in capturing spatial dependencies in taxi trajectories.
• To improve the efficiency and resource utilization of urban taxi services.

Main Methods:

• Construction of an urban topological network by discretizing transportation units and integrating road network nodes.
• Development of a topological trajectory by matching GPS records to the urban topological map.
• Integration of surrounding semantic information to capture spatial dependencies.
• Proposal of a topological graph neural network for spatiotemporal attention calculation.

Main Results:

• The proposed urban topological map significantly enhances trajectory consistency and endpoint certainty.
• The urban topology-encoding spatiotemporal attention network (UTA) model demonstrates improved prediction accuracy.
• The UTA model shows resilience to data sparsity compared to traditional models like HMM, RNN, LSTM, and Transformer.
• Classical models integrated with the proposed urban model showed a ~2% performance increase.

Conclusions:

• The UTA model effectively addresses taxi destination prediction challenges by integrating topological encoding and spatiotemporal attention.
• The proposed urban topological network framework enhances the modeling of taxi trajectories.
• This approach offers a promising solution for optimizing taxi services and reducing urban congestion.