Resource allocation strategy for vehicular communication networks based on multi-agent deep reinforcement learning

Abstract

In complex and high-mobility vehicular communication networks, rapidly changing channel conditions, signal interference, and stringent latency requirements of safety services pose significant challenges to existing wireless resource allocation schemes. We propose a novel resource allocation method named AMADRL. It is based on the multi-agent deep reinforcement learning (MADRL) algorithm and incorporates attention mechanisms (AM). This method first improves the traditional MADRL framework by employing two critic networks to estimate the corresponding global and local reward functions, achieving joint optimization of spectrum and power allocation. This optimization balances the individual interests of agents with the collective benefits, meeting the low-latency communication requirements of vehicle-to-vehicle (V2V) links. And this method effectively reduces the interference to the vehicle-to-infrastructure (V2I) links. Building on this foundation, we further integrate AM into the framework. The AM enables the model to selectively focus on critical information, dynamically adjusting resource allocation strategies. Simulation results demonstrate that, compared with random methods and conventional deep reinforcement learning (DRL) methods, the proposed algorithm exhibits superior convergence speed and stability. It effectively meets the communication requirements of different links and significantly improves spectrum efficiency.