Integrated hybrid algorithm and multi-agent cooperation on computational offloading problem in mobile edge computing

Abstract

Mobile Edge Computing (MEC) has emerged as a promising paradigm to mitigate the limitations of cloud computing, particularly in supporting latency-sensitive and computation-intensive applications in the Internet of Things (IoT) ecosystem. However, efficient resource management in MEC remains a challenge due to device heterogeneity, mobility, and the constrained capacity of edge servers. Computation offloading is one of the techniques used to address the aforementioned challenges. Existing researches uses guided searches, strategic interactions and approximations to make offloading decisions. However, the existing techniques often suffer from slow convergence, high computational time, and an inability to adapt to dynamic network variations, which results in high energy consumption, high latency and scalability issues.  This research addresses the problems by proposing an Integrated Hybrid Algorithm and Multi-Agent Cooperation (IHAMAC). The proposed IHAMAC combines the strengths of model-free Deep Deterministic Policy Gradient (DDPG) with model-based reinforcement learning, enhanced with multi-agent cooperation to improve scalability, energy efficiency, and latency. The hybrid algorithm leverages predictive modeling to improve sample efficiency while retaining robustness through real-environment interactions. The multi-agent framework decentralizes decision-making and adapts to device mobility, mitigating the limitations of centralized control. Extensive simulations demonstrate that IHAMAC outperforms baseline DDPG-PER approaches, achieving lower latency, reduced energy consumption, and better scalability under varying network loads.