Optimal Resource Management for Hierarchical Federated Learning over HetNets with Wireless Energy Transfer

Rami Hamdi, Ahmed Ben Said, Emna Baccour, Aiman Erbad, Amr Mohamed, Mounir Hamdi, Mohsen Guizani

Abstract

Remote monitoring systems analyze environment dynamics in various smart industrial applications, such as occupational health and safety and environmental monitoring. In industrial Internet of Things (IoT) systems, the massive number of devices and performance requirements place significant pressure on computational, network, and energy resources. Distributed training of Machine Learning (ML) and Deep Learning (DL) models for these applications is particularly challenging for resource-limited devices over heterogeneous wireless networks (HetNets). Hierarchical Federated Learning (HFL) addresses this by performing training at multiple layers, offloading tasks to nearby Multi-Access Edge Computing (MEC) units. In this paper, we propose a novel energy-efficient HFL framework enabled by Wireless Energy Transfer (WET) and designed for HetNets with massive Multiple-Input Multiple-Output (MIMO) wireless backhaul. Our approach formulates energy efficiency as a Mixed-Integer Non-Linear Programming (MINLP) problem, optimizing device association and wireless energy management. Due to the high complexity of MINLP, we design a Heuristic Resource Management Algorithm (H2RMA) that respects energy, channel quality, and accuracy constraints with low computational complexity. We also improve network energy consumption through an efficient device scheduling scheme and investigate the impact of device mobility on HFL performance. Extensive experiments confirm the high performance of the proposed approach in terms of training loss and grid energy costs.

Introduction

The proliferation of industrial IoT (IIoT) has enabled sophisticated remote monitoring for occupational safety and environmental health. However, the sheer volume of data generated by thousands of resource-constrained devices—such as wearables and smart sensors—presents a significant bottleneck for traditional centralized AI training. While standard Federated Learning (FL) allows devices to train models locally and share only updates, it often suffers from communication bottlenecks and rapid battery depletion.

To mitigate these issues, Hierarchical Federated Learning (HFL) introduces an intermediate layer of Multi-Access Edge Computing (MEC) units. By aggregating updates locally before communicating with a Central Unit (CU), HFL reduces the burden on the core network. Yet, two primary challenges remain: the extreme energy scarcity of IoT devices and the bias introduced by non-independent and identically distributed (non-I.I.D.) data. This research addresses these hurdles by integrating Wireless Energy Transfer (WET) and a novel resource management strategy.

System Architecture and Methodology

The proposed framework operates over a heterogeneous network where MEC units are equipped with massive MIMO capabilities to handle wireless backhaul to the CU. The architecture leverages three distinct strategies to optimize performance:

1. Wireless Energy Transfer (WET)

To solve the problem of finite battery life, MEC units act as power hubs, wirelessly "beaming" energy to IoT devices. This creates a closed-loop system where the primary constraint shifts from device battery capacity to the cost of grid energy at the MEC level.

2. The H2RMA Algorithm

The core of the resource management strategy is the Heuristic Resource Management Algorithm (H2RMA). The researchers initially formulated the association problem as a Mixed-Integer Non-Linear Programming (MINLP) problem. Because MINLP is NP-hard and unsuitable for real-time industrial applications, H2RMA provides a near-optimal heuristic. Unlike traditional association schemes that prioritize the strongest signal, H2RMA selects MEC-device pairings based on a "divergence-aware" metric. It prioritizes associations that minimize the difference between local data distributions and the global model, ensuring faster and more accurate convergence.

3. Dynamic Device Scheduling and Mobility

To further conserve energy, the system employs a scheduling scheme that deactivates devices requiring excessive energy for a specific training round, provided their absence does not significantly impact model accuracy. Furthermore, to handle the movement of workers across a factory floor, the authors implemented Dynamic HFL Device Association (DHDA). Using Hidden Markov Models (HMMs), the system predicts worker movement states (Static, Normal, or Risky) and dynamically re-associates devices to prevent energy spikes caused by communicating over suboptimal distances.

Experimental Results

The framework was evaluated using the MNIST and CIFAR-10 datasets, distributed in a non-I.I.D. fashion to simulate real-world data skewness.

  • Energy Efficiency: H2RMA significantly outperformed random association baselines. In simulations with 60 devices, the grid energy cost was reduced by several orders of magnitude compared to standard methods.
  • Model Accuracy: The divergence-aware selection allowed the HFL model to reach higher test accuracy (approximately 85-90% on MNIST) more rapidly than traditional schemes.
  • Mobility Resilience: The DHDA approach successfully maintained performance levels even as devices moved, proving that predictive association is vital for dynamic industrial environments.

Conclusion and Future Work

This research demonstrates that combining Hierarchical Federated Learning with Wireless Energy Transfer can effectively sustain large-scale IIoT monitoring systems. By balancing the trade-off between energy consumption and data diversity through the H2RMA algorithm, the framework achieves high accuracy with minimal grid energy expenditure.

Future research will focus on the challenges of maintaining perfect Channel State Information (CSI) in cluttered industrial environments and exploring Reinforcement Learning (RL) to further automate clustering and association strategies in even denser device deployments.

Summary updated

Abstract

Remote monitoring systems analyze the environment dynamics in different smart industrial applications, such as occupational health and safety, and environmental monitoring. Specifically, in industrial Internet of Things (IoT) systems, the huge number of devices and the expected performance put pressure on resources, such as computational, network, and device energy. Distributed training of Machine and Deep Learning (ML/DL) models for intelligent industrial IoT applications is very challenging for resource limited devices over heterogeneous wireless networks (HetNets). Hierarchical Federated Learning (HFL) performs training at multiple layers offloading the tasks to nearby Multi-Access Edge Computing (MEC) units. In this paper, we propose a novel energy-efficient HFL framework enabled by Wireless Energy Transfer (WET) and designed for heterogeneous networks with massive Multiple-Input Multiple-Output (MIMO) wireless backhaul. Our energy-efficiency approach is formulated as a Mixed-Integer Non-Linear Programming (MINLP) problem, where we optimize the HFL device association and manage the wireless transmitted energy. However due to its high complexity, we design a Heuristic Resource Management Algorithm, namely H2RMA, that respects energy, channel quality, and accuracy constraints, while presenting a low computational complexity. We also improve the energy consumption of the network using an efficient device scheduling scheme. Finally, we investigate device mobility and its impact on the HFL performance. Our extensive experiments confirm the high performance of the proposed resource management approach in HFL over HetNets, in terms of training loss and grid energy costs.
BibTeX
@article{hamdi2023optimal,
  author = {Rami Hamdi and Ahmed Ben Said and Emna Baccour and Aiman Erbad and Amr Mohamed and Mounir Hamdi and Mohsen Guizani},
  title = {{Optimal Resource Management for Hierarchical Federated Learning over HetNets with Wireless Energy Transfer}},
  journal = {IEEE Internet of Things Journal},
  year = {2023},
  url = {https://openreview.net/forum?id=PHsyOpiPSk},
}
Figure