Multi-objective federated learning traffic prediction in vehicular network for intelligent transportation system

Area of Science:

• Transportation Science
• Artificial Intelligence
• Data Science

Background:

• Accurate spatial-temporal prediction of future freight traffic speed is crucial for effective metropolitan traffic management.
• Current methods often rely on centralized data processing, posing challenges for privacy and efficiency.

Purpose of the Study:

• To develop a novel approach for freight traffic speed prediction using multi-objective federated learning.
• To enhance traffic management strategies through improved spatial-temporal forecasting.

Main Methods:

• Implemented a collaborative training approach among participants, avoiding a centralized cloud server.
• Utilized reinforcement learning for dynamic decision-making and graphical models for traffic intensity identification.
• Integrated federated learning with multi-objective optimization for accurate traffic pattern forecasting.

Main Results:

• Achieved a prediction accuracy of 97.45% for traffic speed.
• Demonstrated a communication delay of 23.4% and a packet delivery ratio (PDR) of 92.45%.
• Showcased effective capture of interconnections between metropolitan areas and neighboring cities through visualization.

Conclusions:

• The proposed multi-objective federated learning approach significantly outperforms existing methods for freight traffic speed estimation.
• The method offers improved communication efficiency, packet delivery, and resource utilization.
• This approach provides a robust framework for understanding and managing complex metropolitan traffic dynamics.