An IPSO-assisted DNN operation scheme in UAV-mounted MEC networks

Abstract

With the growth of the Internet in recent years, applications have higher and higher requirements on the computation capacity of User Equipments (UEs). Offloading compute-intensive tasks to edge servers becomes a promising solution. Therefore, Unmanned Aerial Vehicles (UAVs)-mounted Mobile Edge Computing (MEC) emerges as the times require, with the ability to provide task offloading for ground UEs while providing short-term network services for emergencies. However, due to the changing environment, how to provide fast decision solutions remains a hot topic of research. In this paper, we propose an Improved Particle Swarm Optimization (IPSO)-assisted Deep Neural Network (DNN) operation scheme, which combines a heuristic algorithm with neural networks to provide an efficient operation scheme. We consider the need to shorten UE latency and reduce UE energy consumption while ensuring fairness in offloading, which is constructed as a Mixed Integer NonLinear Programming (MINLP) problem. We solve it by the IPSO algorithm and provide high-quality labeled data for training the neural network. The runtime leverages the trained neural network to make fast decisions. Finally, simulation experiments show that our scheme has some superiority in adapting quickly to changing environments.