Federated Synergy: Hierarchical Multi-Agent Learning for Sustainable Edge Computing in IIoT

The Industrial Internet of Things (IIoT) presents significant challenges in task offloading, resource allocation, and energy efficiency, necessitating intelligent, scalable, and adaptive solutions. This paper introduces the Hierarchical Multi-Agent Federated Actor-Critic (HMAFAC) algorithm, a novel approach that integrates hierarchical reinforcement learning with multi-agent federated learning to enhance decision-making in dynamic edge computing environments. It employs multi-agent reinforcement learning to optimize resource allocation and task offloading by leveraging local Actor-Critic models on IoT edge devices. A Deep Q-Network (DQN) agent manages task offloading decisions based on real-time system states, ensuring efficient task execution. To maintain data privacy and reduce communication overhead, federated learning aggregates model updates at the edge server, eliminating the need for raw data exchange. Additionally, a hierarchical framework enables adaptive scheduling, allowing flexible computational scaling from edge to cloud servers based on task complexity and network conditions. By effectively reducing latency and energy consumption, enhancing system adaptability, and maintaining computational efficiency, HMAFAC overcomes key limitations of existing IIoT frameworks. Simulation results validate its superior performance compared to baseline methods regarding resource utilization, energy conservation, and model convergence. These findings establish HMAFAC as a robust and scalable framework for optimizing IIoT applications in distributed and resource-constrained environments. Future work will explore advanced federated learning techniques, enhanced privacy mechanisms, and real-world deployment in industrial systems to improve efficiency and scalability further.