An Ensemble Deep Reinforcement Learning-Based Asynchronous Scheduling Mechanism for 5G-TSN

Integrating Time-Sensitive Networking (TSN) with 5G enables Industrial IoT (IIoT) systems to achieve low latency and high reliability. However, deterministic cross-domain scheduling remains challenging due to strict synchronization demands and wireless channel variations. This paper proposes an asynchronous scheduling mechanism that consists of Ensemble Deep Reinforcement Learning and Online Asynchronous Flow Scheduling. The former integrates three reinforcement learning agents to generate a near-optimal Gate Control List (GCL) in complex asynchronous network environments, and the latter reallocates time-slot paths online based on the GCL to make fuller use of idle slot resources. Extensive OMNeT++ simulations show that the proposed asynchronous scheduling mechanism efficiently utilizes network resources, reduces delay from wireless fluctuations and asynchrony, and improves throughput.