Time Synchronization-Aware Edge-End Collaborative Network Routing Management for FL-Assisted Distributed Energy Scheduling

Federated learning (FL)-assisted model training plays an important role in distributed energy scheduling of smart park. However, the time synchronization error between edge and end sides and the adversarial routing competition cause poor accuracy and high delay of model training. In this paper, we address this challenge and propose a time synchronization-aware edge-end collaborative deep Q network-based routing management algorithm named TSA-RM. TSA-RM minimizes the weighted sum of model training loss function and delay via routing optimization. TSA-RM achieves time synchronization awareness and avoids adversarial competition by incorporating time synchronization related information in state space construction and relay selection related information in penalty function design. Simulation results verify the superior performance of TSA-RM in terms of global loss function, model training delay, and time synchronization error compared with two state-of-the-art algorithms.