Deep Reinforcement Learning for Edge Caching with Mobility Prediction in Vehicular Networks

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Connected vehicles generate significant data, necessitating efficient content delivery.
• Centralized servers face latency issues with increasing vehicle data requests.
• Mobile Edge Computing (MEC) offers a solution by bringing computation closer to users.

Purpose of the Study:

• To propose an optimized caching strategy for Roadside Units (RSUs) and On-Board Units (OBUs) in MEC environments.
• To maximize caching node throughput while satisfying stringent delay constraints for connected vehicle services.
• To address the challenges of dynamic content popularity and vehicle mobility in caching.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) networks to predict vehicle trajectories based on temporal and spatial data.
• Employed Deep Deterministic Policy Gradient (DDPG) algorithms to dynamically manage content caching sizes based on popularity.
• Implemented a dual caching approach involving both RSUs and vehicles as caching nodes.

Main Results:

• The proposed caching strategy significantly improved caching node throughput compared to existing methods.
• Demonstrated enhanced satisfaction of delay constraints in diverse experimental environments.
• Showcased a reduction in the overall cost associated with content updates.

Conclusions:

• The integration of predictive vehicle location and dynamic content popularity management is effective for MEC caching.
• The proposed LSTM and DDPG-based approach offers a robust solution for optimizing connected vehicle services.
• This strategy provides a scalable and efficient method for enhancing user experience in vehicular networks.