MEC-Enabled Hierarchical Federated Learning for Resource-Aware Device Selection in IIoT

Area of Science:

• Artificial Intelligence
• Computer Science
• Electrical Engineering

Background:

• Hierarchical Federated Learning (HFL) with Mobile Edge Computing (MEC) is promising for Industrial Internet of Things (IIoT) due to reduced communication overhead.
• Dynamic device participation and varied training objectives in real-world IIoT hinder model convergence and system performance.

Purpose of the Study:

• To propose a dynamic device selection strategy for HFL in IIoT to enhance model convergence stability.
• To optimize system resource consumption and model performance under dynamic conditions.

Main Methods:

• A device selection strategy based on task completion probability was developed for dynamic device participation.
• An optimization objective was formulated to minimize the loss function under resource constraints.
• The objective was reformulated as a loss upper bound minimization problem and solved iteratively.

Main Results:

• The proposed method demonstrated superior resource efficiency and training stability in simulations.
• Compared to state-of-the-art HFL, the new method reduced average training delay by 18% and energy consumption by 22%.
• Competitive model accuracy was maintained under dynamic IIoT conditions.

Conclusions:

• The joint optimization strategy effectively addresses challenges in dynamic HFL for IIoT.
• The approach validates the feasibility of balancing resource efficiency, training stability, and model performance in practical IIoT applications.