Deep Reinforcement Learning Based Intelligent Resource Allocation in Hybrid Vehicle Scenario

In recent years, there has been rapid development in vehicular networks and autonomous driving. While vehicles of various intelligence levels are becoming more common on the road, most research overlooks the data distribution across different vehicles in multicast scenarios. Our aim is to allow different kinds of vehicles to receive the needed content in a multicast scenario and to fulfill certain freshness requirements. Although deep reinforcement learning (DRL) has been widely used to address this issue, it suffers from slow training convergence and unstable performance. Hence, this study proposes combining DRL algorithms with behavior cloning and action mask, leveraging prior knowledge and expert algorithms to enhance performance. Finally, the freshness of the data content is ensured for all kinds of vehicles and effective data transmission is achieved. The simulation results indicate a significant improvement in the training efficiency and performance in our proposed method, with 15.6% to 31.9% improvement in terms of effective traffic compared to other counterparts.