A Multi-Agent Deep Reinforcement Learning-Based Popular Content Distribution Scheme in Vehicular Networks

Area of Science:

• Vehicular Networks
• Artificial Intelligence
• Communication Systems

Background:

• The Internet of Vehicles (IoV) relies on vehicle-to-everything (V2X) for data services.
• Popular Content Distribution (PCD) in IoV faces challenges due to vehicle mobility and roadside unit (RSU) limitations.
• Vehicle-to-vehicle (V2V) communication collaboration is crucial for efficient content delivery.

Purpose of the Study:

• To propose a novel Multi-Agent Deep Reinforcement Learning (MADRL)-based scheme for Popular Content Distribution (PCD) in vehicular networks.
• To enhance the efficiency and reduce transmission delay in delivering popular content to vehicles.
• To address the challenges posed by vehicle mobility and RSU coverage limitations in IoV.

Main Methods:

• A MADRL-based scheme is developed, with each vehicle acting as an MADRL agent.
• Vehicles are clustered using spectral clustering to manage complexity during V2V communication.
• The Multi-Agent Proximal Policy Optimization (MAPPO) algorithm trains the agents, incorporating self-attention and invalid action masking.

Main Results:

• The proposed MADRL-PCD scheme significantly improves PCD efficiency compared to existing methods.
• The scheme achieves a notable reduction in transmission delay for popular content.
• Experimental results validate the effectiveness of the MADRL-PCD approach over coalition game and greedy strategies.

Conclusions:

• The MADRL-PCD scheme offers a superior solution for popular content distribution in vehicular networks.
• The integration of spectral clustering, self-attention, and MAPPO effectively optimizes content delivery.
• This research contributes to more efficient and faster data dissemination in the Internet of Vehicles.