Multi-mode Task Offloading in Digital Twin Edge Network Assisted Industrial Internet of Things

In the context of the Industrial Internet of Things (IIoT), the task offloading decisions in real-time industrial production are effectively supported by the Digital Twin Edge Network (DITEN), leveraging its capabilities for virtual-physical collaboration and low-latency. However, the single-hop offloading mode fails to fully utilize the computing resources of edge servers, posing challenges in meeting the low-latency requirements of industrial devices. To tackle this issue, a multi-mode task offloading scheme is proposed based on the task offloading framework of DITEN within IIoT. This scheme encompasses local computing, single-hop offloading, and multi-hop offloading modes, aiming to enhance resource utilization and reduce decision-making risks. After thoroughly defining the necessary models, the problem of minimizing task processing delay for industrial devices is transformed into a Markov Decision Process. Subsequently, an optimization algorithm is designed, which integrates the Multi-Agent Proximal Policy Optimization algorithm with a temperature-controlled Softmax function to facilitate policy exploration. Experimental results demonstrate that the proposed algorithm outperforms comparative algorithms in terms of both reward and average task completion delay.