Edge-Oriented Resource Scheduling Algorithm Based on Deep Reinforcement Learning

In the past few decades, the rapid development of information technology and the continuous innovation of computing paradigm have ushered people into a new era of edge computing. Compared to traditional cloud computing, edge computing provides better performance by bringing computing and storage capabilities to edge of network. As the scale of edge networks expands, the increasing number of latency-sensitive and compute-intensive tasks has led to a growing demand for computational power in terminal devices. Proper resource scheduling policies play a decisive role in improving the performance of edge networks. For the edge side resource scheduling scenario, this paper studies the multi-agent Counterfactual Soft self-Attention Actor-Critic algorithm (mCSAAC). To address the issue of credit assignment where global rewards during centralized training fail to reflect individual contributions, a counterfactual policy gradient is proposed to enhance the overall performance of the network. Furthermore, to tackle the short-sightedness problem introduced by fully decentralized execution, a self-attention communication mechanism is introduced between each agent, facilitating communication and reducing information loss in the environment. Through simulation experiments, the effectiveness of the proposed algorithm in reducing system energy consumption and task execution latency has been validated.