A Distributed Adaptive QoS-Aware TSCH Scheduling to Support Heterogeneous Traffic in IIoT Using Fuzzy Reinforcement Learning

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

The Industrial Internet of Things (IIoT), a pillar of Industry 4.0, collects environmental data via Low-Power and Lossy Networks (LLNs) and employs Time Slotted Channel Hopping (TSCH) to schedule communications among LLN nodes. Designing an efficient TSCH scheduler is challenging due to resource limitations, scalability concerns, environmental dynamics, traffic heterogeneity, and stringent Quality of Service (QoS) requirements. Various TSCH scheduling approaches leverage Reinforcement Learning (RL) to enable a distributed, dynamic, self-learning method that fine-tunes scheduling without prior knowledge of the environment. However, the main drawbacks of these approaches lie in their inability to support heterogeneous traffic QoS requirements and interpret scheduling states. Consequently, these algorithms often employ a best-effort strategy, which is prone to issues such as state-space explosion and slow convergence. In this paper, we propose a Distributed, Adaptive, and QoS-aware (DAQ) approach that support heterogeneous traffic using a hybrid design based on Fuzzy Rule-Based System (FRBS) and RL. The FRBS component, empowered by granular computing, discretizes large and continuous state spaces into a limited number of manageable states, thereby reducing state-space complexity and accelerating RL convergence. These states model TSCH scheduling as a Markov Decision Process, and RL component attempts to discover an optimal scheduling policy. Evaluation results indicate that, compared to existing approaches, the DAQ maintains performance for Non-Real-Time traffic while achieving significant improvements for Real-Time traffic. Specifically, it improves reliability, average delay, and maximum delay by up to 7%, 44%, and 43%, respectively, and reduces energy consumption by up to 44%.

Article activity feed