Eecs-fl: energy-efficient client selection for federated learning in AIoT

(2025) 2025:12 Zhang et al. J Wireless Com Network https://doi.org/10.1186/s13638-025-02435-2 EURASIP Journal on Wireless Communications and Networking Open Access RESEARCH Eecs‑fl: energy‑efficient client selection for federated learning in AIoT Yiyang Zhang1, Yiming Luo1, Tao Yang1*, Xiaofeng Wu1 and Bo Hu1 *Correspondence: 1 Department of Electronics Engineering, School of Information Science and Technology, Fudan University, Shanghai, China Abstract The Artificial Intelligence of Things (AIoT) ecosystem faces significant challenges related to limited client energy budgets and resource heterogeneity, particularly when employing the Federated Learning (FL) framework. This paper presents a novel energy-efficient client selection algorithm for FL, designed to address these challenges by integrating Wireless Power Transfer (WPT), where WPT involves in the client selection optimization, based on real-time energy availability and resource heterogeneity. We formulate the client selection problem as a multi-dimensional knapsack problem (MKP) and solve it using dynamic programming to maximize energy efficiency while maintaining fast convergence. Experimental results show that incorporating WPT leads to a reduction in unit energy consumption by over 24.54%; while, the proposed algorithm achieves a reduction of over 15.31% compared to random selection. The proposed approach improves energy utilization, demonstrates strong resilience to client heterogeneity, and adapts efficiently to varying energy supply conditions. Keywords: Artificial intelligence of things (AIoT), Federated learning (FL), Client selection, Wireless power transfer (WPT), Energy-efficient 1 Introduction The Internet of Things (IoT) is rapidly transforming traditional networks by connecting a vast number of physical devices and enabling ubiquitous sensing and computing capabilities [1]. As a result, the integration of Artificial Intelligence (AI) with IoT has led to the emergence of the Artificial Intelligence of Things (AIoT), realizing the concept of an “Internet of Everything.” AIoT enhances IoT systems with advanced perception, learning, reasoning, and action capabilities, primarily driven by sophisticated data analysis and processing [2]. Traditionally, AIoT systems adopt a centralized AI approach, where data integration and model training occur in centralized data centers or on individual clients [3]. Federated Learning (FL) offers a significant paradigm shift by decentralizing model training and transmitting only model parameters rather than raw data. This shift reduces the burden on centralized systems and enhances privacy protection, making FL a critical solution in sectors such as healthcare, transportation, and industry [4–9]. In this context, FL enables the AIoT ecosystem to continue expanding efficiently and securely. © The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/. Zhang et al. J Wireless Com Network (2025) 2025:12 However, the growing complexity of AIoT environments presents significant challenges for FL, including client mobility, resource heterogeneity, and energy limitations, which hinder global training participation and convergence [10–12]. Specially, the limited battery power pose potentially the threat of disrupting model training and even breakdown the seamless operation of AloT systems. To counter these challenges, client selection algorithms have been proposed to optimize resource utilization in FL [13–21]. Chen et al. [13] introduced asynchronous FL by enhancing server aggregation; while, Wu et al. [14] implemented a semi-asynchronous scheme to reduce client waiting times and improve efficiency. However, these methods overlook client heterogeneity in resource distribution. Nishio et al. [15] advanced the field by proposing a client selection mechanism designed to maximize the number of clients participating in the aggregation process within a limited time. This approach ensures energy-efficient training by increasing client-side engagement. Building on this foundation, subsequent research [16–18] has refined client selection to balance convergence speed and the challenges posed by data heterogeneity. However, these approaches remain constrained by resource and energy limitations. Nonetheless, existing methods lack mechanisms to dynamically adapt to real-time resource and energy availability [22], particularly in heterogeneous AIoT environments. Additionally, while approaches such as Wireless Power Transfer (WPT) [23] and Radio Frequency Energy Harvesting (RFEH) [24] have been proposed to mitigate energy constraints, the integration of reliable energy harvesting methods with FL frameworks remains underdeveloped, requiring further research to establish practical and effective solutions [25]. The introduction of WPT brings new challenges, as the arrival and consumption of energy make the system more complex, and existing algorithms struggle to effectively address these complexities. These limitations form critical research gaps that motivate the present study. To address these gaps, our paper introduces an energy-efficient client selection algorithm for FL in AIoT. This algorithm not only addresses convergence and efficiency issues but also adapts to client resource heterogeneity, offering a robust solution for energy-limited and data-intensive AIoT environments. The proposed solution leverages real-time energy profiling and incorporates dynamic allocation strategies to enhance the integration of energy harvesting methods, ensuring sustained client participation and seamless operation under strict energy constraints. Unlike traditional methods, the algorithm actively considers the diverse energy and computational profiles of AIoT clients, providing a more tailored and robust selection process that maximizes energy efficiency while maintaining rapid convergence. The main contributions of our research are: 1. Beacon-WPT Mechanism for Energy Assurance: We propose a novel Beacon-WPT mechanism that for (...truncated)