Scholarlink

<p>Effective communication scheduling in TSCH-based IoT networks is critical for meeting the reliability, latency, and energy requirements of industrial applications. Existing schedulers – whether static, heuristic, or centralized – fail to adapt rapidly to dynamic traffic conditions, leaving an unresolved gap in lightweight, fully decentralized scheduling for constrained 6LowPAN devices. In this paper, we propose a decentralized Q-learning-based TSCH scheduler where each node acts as an independent reinforcement learning agent. At every slotframe boundary, each agent selects the number of shared communication and sleep slots based on a unified reward combining link quality (ETX) and local traffic load, without requiring inter-node coordination. We introduce a slotframe-aligned reward formulation, a sensitivity-validated coefficient analysis, and a randomized Q-table initialization combined with an <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\varepsilon \)</EquationSource> </InlineEquation>-greedy exploration strategy to prevent premature convergence. Experiments conducted under the Cooja network emulator with the Multi-path Ray-Tracer Medium (MRM) propagation model, across random topologies of 20 and 30 nodes, show that the proposed scheduler consistently achieves the highest PDR (up to <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(99.68 \pm 0.38\)</EquationSource> </InlineEquation>%) and the lowest duty cycle (down to <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(0.08 \pm 0.005\)</EquationSource> </InlineEquation>%) across all traffic loads (1–30&#xa0;pkt/min), while maintaining end-to-end latency below 200&#xa0;ms – representing up to 95% latency reduction compared to baseline schedulers. The network converges in approximately 11&#xa0;s, independent of network density, confirming the scalability of the proposed decentralized design.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Actions oriented scheduling based on reinforcement Q-learning: an optimized allocation of time-frequency communications for TSCH in IoT networks

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Abdelhadi El Oudrhiri Hassani,
  • Ilham El Mourabit,
  • Adil Salbi,
  • Aicha Sahel,
  • Abdelmajid Badri

摘要

Effective communication scheduling in TSCH-based IoT networks is critical for meeting the reliability, latency, and energy requirements of industrial applications. Existing schedulers – whether static, heuristic, or centralized – fail to adapt rapidly to dynamic traffic conditions, leaving an unresolved gap in lightweight, fully decentralized scheduling for constrained 6LowPAN devices. In this paper, we propose a decentralized Q-learning-based TSCH scheduler where each node acts as an independent reinforcement learning agent. At every slotframe boundary, each agent selects the number of shared communication and sleep slots based on a unified reward combining link quality (ETX) and local traffic load, without requiring inter-node coordination. We introduce a slotframe-aligned reward formulation, a sensitivity-validated coefficient analysis, and a randomized Q-table initialization combined with an \(\varepsilon \) -greedy exploration strategy to prevent premature convergence. Experiments conducted under the Cooja network emulator with the Multi-path Ray-Tracer Medium (MRM) propagation model, across random topologies of 20 and 30 nodes, show that the proposed scheduler consistently achieves the highest PDR (up to \(99.68 \pm 0.38\) %) and the lowest duty cycle (down to \(0.08 \pm 0.005\) %) across all traffic loads (1–30 pkt/min), while maintaining end-to-end latency below 200 ms – representing up to 95% latency reduction compared to baseline schedulers. The network converges in approximately 11 s, independent of network density, confirming the scalability of the proposed decentralized design.