A Distributed Task Offloading Strategy Based on Multi-Agent Systems in Mobile Edge Computing

With the rapid development of smart applications and mobile communication technologies, mobile edge computing (MEC) has made significant progress. However, the increase in the number of terminal devices and the growing complexity of service scenarios pose new challenges to efficient task processing. To address this challenge, this paper investigates the dynamic task offloading problem in the “cloud-edge-device” three-layer edge computing network architecture. To address this problem, we propose a distributed collaboration mechanism that supports device-level collaboration, device-edge collaboration, and edge-level collaboration, and design a Multi-Agent-based Task Offloading Algorithm (MATOA). In this algorithm, each agent interacts with each other through a deep reinforcement learning algorithm in a dynamic network environment to achieve continuous learning and parameter optimization. Simulation experiments show that compared with the baseline algorithm, this approach has good robustness in dynamic network environments, and can effectively reduce the overall execution latency of the task during the peak period of the task.