RESOURCE ALLOCATION OPTIMIZATION FOR EFFECTIVE VEHICLE NETWORK COMMUNICATIONS USING MULTI-AGENT DEEP REINFORCEMENT LEARNING

In the realm of vehicular networks, ensuring top-notch services for high-speed mobile vehicles remains a formidable challenge. This study squarely addresses the spectrum-sharing predicament within the Internet of Vehicles (IoV) by scrutinizing the spectrum shared by V2I links serving multiple V2V links. The swiftness of vehicular movement ushers in challenges for centralized resource management, necessitating an innovative approach. To surmount this obstacle, the resource allocation is casted as a Multi-agent Deep Reinforcement Learning (RL) and a distributed Multi-Agent Deep Deterministic Policy Algorithm (MADDPG) algorithm is introduced. Each agent autonomously interacts with the vehicular network, assessing its local state and receiving a shared reward. Through collaboration, the Critic network refines power control for each agent. The reward function and training mechanism empower the multi-agent algorithm to attain distributed resource allocation, markedly enhancing both V2I link capacity and V2V transmission rates. This pioneering approach optimizes spectrum utilization, ushering in superior service quality for high-speed mobile vehicles in vehicular networks.