Intelligent Online Computation Offloading for Wireless Powered Vehicle Edge Computing

In Internet of Vehicles, task processing performance of vehicle is always limited by its finite energy and low computing power. Wireless Powered Vehicular Edge Computing (WP-VEC) is a solution to this limitation, where vehicles can harvest energy and offload tasks. However, many related works pay less attention to timeliness of offloading strategy, and thus suffer poor adaptability to highly dynamic environments. In this paper, we consider a WP-VEC network with binary offloading decision. We jointly optimize offloading decision and resource allocation to minimize weighted sum of delay. Here, the resource allocation includes time allocation between energy harvesting phase and offloading phase, and communication and computation resource allocation in offloading phase. This problem is formulated as a Mixed Integer Programming (MIP) problem, which is hardly to be solved within channel coherence time by traditional methods. To overcome this, we utilize deep neural network and optimization technologies to design such an Intelligent Online Computation Offloading (IOCO) algorithm. Simulation results verify its excellent performance on effectiveness, convergence, and robustness, which verifies IOCO can adapt to highly dynamic environments.