Resource Allocation in Vehicular Networks based on Hybrid Proximal Policy Optimization

With rapid development of intelligent vehicles, radio resource allocation in cellular D2D-based V2X communication (C-V2X) has become a key area of research. Traditional resource allocation methods face challenges as difficulty in optimizing complex problems and high computational complexity. While deep reinforcement learning (DRL) has recently emerged as a promising solution, existing DRL-based algorithms exhibit inherent limitations, including quantization errors from discretized power levels and costly retraining associated with weightpredefined optimization. This study introduces an innovative framework to address these challenges. For parameterized action spaces, this paper introduces a multi-agent hybrid policy proximal optimization (MAHPPO) algorithm for continuous power control and spectrum allocation in the cooperative <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{C}-\mathrm{V} 2 \mathrm{X}$</tex> networks, eliminating quantization errors. Furthermore, we develop a parameter transfer-based method to efficiently compute the Pareto Front, significantly reducing resource consumption linked to dynamic weight adjustments. Simulation results demonstrate that the proposed method ultimately yields favorable resource allocation outcomes, enhancing user service quality.