作者：Sinan Pan, Lei Shi, Han Wu, Yingying Chen, Huaili Liu, Hao Xu发表刊物：CollaborateCom 2024

年份：August 2025

摘要：Clustered Federated Learning (CFL) based on the edge computing (EC) environment has great potential to promote the realization of artificial intelligence at the network edge. However, due to the large network area and the limited energy of edge devices, excessive energy consumption may occur if all devices participate in parameter transmission after training. This can result in insufficient energy for some devices, making it difficult to complete the training tasks. In this paper, we aim to reduce client energy consumption and total transmission energy consumption by the model partition technique and the adjustment of transmission modes among clients. Firstly, we establish the mathematical model, and show that the transmission energy consumption can be effectively reduced by optimizing the transmission distance. Secondly, we design the Distance Best K-Means (DBKM) algorithm to obtain the optimal transmission distance among clients. Finally, we propose a method to dynamically adjust the client training mode based on the client’s energy. Simulation results show that compared with the other three methods, our algorithm can reduce the transmission energy consumption by 27%–82%.

参考文献拷贝字段：Sinan Pan, Lei Shi, Han Wu, Yingying Chen, Huaili Liu, Hao Xu. Client-Oriented Energy Optimization in Clustered Federated Learning with Model Partition[C]. 19th EAI International Conference on Collaborative Computing: Networking, Applications and Worksharing (CollaborateCom), Wuzhen, China, November 15-17, 2024: 195-213. DOI:https://doi.org/10.1007/978-3-031-54531-3_15