A DNN Operation Scheme Based on IPSO for UAV-Assisted MEC Networks

Applications' demands on computing resources have increased in recent years. One potential option is to offload tasks with high computational demands to edge servers. Therefore, Mobile Edge Computing (MEC) deployed on Unmanned Aerial Vehicles (UAVs) emerges as the need arises to offer task offloading for ground User Devices (UDs). However, due to the changing environment, how to offer quick decision-making solutions is still a popular research topic. In this paper, we provide a Deep Neural Network (DNN) operation scheme based on Improved Particle Swarm Optimization (IPSO), which combines a conventional heuristic approach with neural networks to give an effective operation scheme. We take into account the requirement to decrease latency and conserve energy while maintaining equity in offloading, which is formulated as a Mixed Integer Non-Linear Programming (MINLP) issue. We first use the IPSO approach to generate labeled data and then train the neural network. The well-trained neural network is utilized to generate quick decisions. Finally, simulation results reveal our scheme's advantage in terms of fast environment adaptation.