面向空中联邦学习的边缘智能感知模型优化方法研究OA

Integrated sensing and communication,empowered by artificial intelligence,endows 6G networks with high-precision sensing capabilities of the physical world.To efficiently process the massive sensing data collected at edge nodes and reduce model training delay,aerial federated learning has emerged.This method leverages the superposition property of wireless channels to synchronize model parameter updates across multiple nodes.However,the specific impact of wireless channel noise on model performance in aerial federated learning necessitates in-depth investigation.This study aims to optimize model generalization capabilities,incorporating delay and energy consumption constraints to develop a joint optimization framework for sensing and communication resources.The transmission power allocation problem is addressed using a grid search method,exploring the impact of data batch size and channel noise on model generalization capabilities.Simulation experiments demonstrate that the proposed aerial federated learning strategy effectively enhances the model's generalization performance.