Abstract

The rapid development of IoT services has led to an increasing need for efficient resource scheduling at MEC in terms of both latency-sensitive and energy-limited environments. However, conventional optimization strategies can't effectively adapt to dynamically changing networks, heterogeneous devices, and the multiplicity of optimization objectives. This paper have presented a unified hybrid MEC system model integrated by DRL, GAT, MARL, QIO and DT to overcome these limitations, enabling intelligent adaptive task offloading, predictive workload control, and economically aware resource allocation in large scale MEC-IoT networks. The simulations have shown significant improvement of the proposed model over baseline and state-of-the-art techniques. Through simulations, this study have achieved low latency of 80ms, 110ms and 140ms for 50, 100 and 150 devices which is 20-30% lower than that of the second-best technique respectively. The system consumes less energy (0.60 J, 0.85 J and 1.10 J respectively) compared to the other methods. Fairness of devices (0.97, 0.96 and 0.95 respectively) shows fair allocation. Ablation study indicates that all individual modules (DRL for adaptive decisions, GAT for dependencies among network topology, MARL for equilibrium-based pricing, QIO for broader and efficient global exploration, and DT for predictive optimization) play an indispensable and complementary role. The outcomes exhibit great scalability, efficiency and adaptability of the framework to the MEC-IoT environment. The future work would involve deploying the model on real testbeds, integrating privacy and security-preserving solutions and extending the model to heterogeneous edge computing and 6G networks. This proposed hybrid MEC system would prove to be intelligent, adaptive and efficient framework for next generation resource scheduling on MEC-enabled IoT in many scenarios, such as smart cities, industrial IoT and autonomous vehicles.