Cybertwin-Driven Multi-Intelligent Reflecting Surfaces aided Vehicular Edge Computing Leveraged by Deep Reinforcement Learning

Abstract

Recently, the cybertwin-driven intelligent internet of vehicles has received widespread consideration in modern smart cities which makes it possible to run high dimensional, low-latency tolerating, and computational-intensive tasks on the vehicles. Thanks to the development in mobile edge computing, the so-called vehicular edge computing allows mobile vehicles to offload their tasks to the road-side unit or hybrid access point due to the limited computation capability. In this paper, we consider a cybertwin-driven internet of vehicle system that provides computing services for mobile vehicles in local area network or wide area network aided with multi-intelligent reflecting surfaces. Based on this system model, we investigate an optimization problem to jointly maximize the sum of data rate in wide area network, and the sum of energy utilities of vehicles. However, in the proposed system model, it is complicated to design the optimal phase, scheduling and offloading decision policy. To solve this issue, we propose a block coordinate descent and deep reinforcement learning based intelligent IoV computing policy. Numerical results have verified that the proposed algorithm can achieve better IoV computing performance compared with four relative benchmark algorithms.