DRL based offloading of industrial IoT applications in wireless powered mobile edge computing

Abstract

Mobile edge computing is the network technology for providing computing resources of edge computing server to Internet of Things (IoT) applications. Additionally, wireless power transfer (WPT) technology can provide stable energy supplying to IIoT nodes and to overcome the limited node lifetime problem faced when using batteries. In this paper, the wireless powered mobile edge computing network is considered where the edge computing server transfers RF energy to IIoT nodes which use harvested energy to offload partial computation workload based on OFDMA and also conduct local computation. The aim is to maximise the weighted sum computation rate by jointly optimising the WPT duration and the amount of energy used for offloading at each node for each time frame. This paper proposes an offloading approach based on deep reinforcement learning which is able to quickly obtain the near-optimal offloading solutions. Specifically, the original offloading problem is decomposed into the sub-problem of optimising the energy allocated for offloading under a given WPT duration and the top-problem of optimising the WPT duration. Simulation results demonstrate that the proposed algorithm can achieve the near-optimal weighted sum computation rate with very low complexity, which is tailored for the practical dynamic-channel environment.