Cost-Efficient 6G Space-Air-Ground Integrated Mobile Edge Computing for Smart City: A PPO-Based Offloading Decision and Resource Allocation Algorithm
Wenwu Zhu,Xin Chen,Libo Jiao,Geyong Min,Wang Li
DOI: https://doi.org/10.1109/hpcc-dss-smartcity-dependsys60770.2023.00041
2023-01-01
Abstract:With the joint progress of society and technology, smart cities are emerging in a new light, seamlessly integrating digitization with urban development, leading modern society towards a smarter and more sustainable future. However, the construction of smart cities faces numerous challenges, including issues such as large-scale connectivity, low-latency communication, immediate user task responsiveness, and high energy consumption. The Space-air-ground integrated networks (SAGIN), serving as the central focus of sixth generation mobile communications (6G) technology development, amalgamate ground, aerial, and space communication resources. It is seen as an effective solution for smart city development. In this paper, we propose a solution that combines the SAGIN with mobile edge computing (MEC), aiming to optimize system efficiency. Specifically, we introduce a three-tier heterogeneous network architecture. Within this framework, user devices (UDs) and unmanned aerial vehicles (UAV s) generate computational tasks, which can be processed locally or offloaded to UAV s or satellites for computation. Subsequently, we established the SAGIN system model. Within this model, we individually modeled the energy harvesting (EH) process, offloading paths, communication links, latency, and energy consumption, and provided a formalized description of the optimization objectives. Finally, we introduced an offloading decision and resource allocation algorithm for SAGIN system based on proximal policy optimization (ODRA-PPO) and employed the proposed ODRA-PPO algorithm to optimize computational resources and offloading decisions. Through extensive comparative experiments, we demonstrated that the proposed approach significantly reduces system cost compared to conventional methods, while simultaneously meeting user task requirements.