Private Edge Computing Resource Allocation and Communication Optimization Based on Federated Learning

Private Edge Computing (PEC) in IoT extends cloud services to the network edge to enhance applications. PEC faces challenges such as device heterogeneity, bandwidth constraints, and high latency. In this paper, we propose an endedge collaborative clustered edge computing using stochastic drift federated optimization (SDFO). SDFO uses federated learning to train models locally in private isomorphic clusters, leveraging the power of edge devices to solve the device and data heterogeneity (Non-IID) problem and reduce reliance on central servers. SDFO optimizes the particle swarm algorithm through iterative optimization by transmitting only the relevant training parameters to reduce communication costs. Experimental results on CIFAR-10 and MNIST datasets validate the effectiveness of SDFO in handling non-IID data, reducing cost, and improving convergence and accuracy.