Scholarlink

<p>The rapid growth of 6G networks has intensified the need for ultra-reliable low-latency communication (URLLC) and intelligent task offloading in multi-edge computing environments. This study proposes an enhanced Hybrid Double Deep Q-Network–Fuzzy SARSA (DDQN–Fuzzy SARSA) model integrating experience replay memory to achieve stable, interpretable, and real-time offloading decisions. The model combines the long-term action-value optimization of DDQN with the adaptive rule-based reasoning of Fuzzy SARSA (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\lambda\)</EquationSource> </InlineEquation>), while a task-failure-aware reward function incorporating Task Failure Rate (TFR) ensures rapid avoidance of congested edge nodes and reinforces QoS guarantees. A comprehensive system model considering MAN/WAN delays, VM utilisation, server coordinates, and heterogeneous IoT workloads was developed and evaluated through extensive MATLAB simulations. Results demonstrate that the proposed framework significantly surpasses standalone DQN, Fuzzy SARSA, PPO, A3C, and PSO baselines, achieving 0.35&#xa0;ms latency, 99.9999% reliability, 99.67% energy efficiency, and a 40% reduction in task failure rate. The hybrid model also generalises effectively across diverse applications including healthcare, AR/VR, infotainment, and compute-intensive services. Overall, the proposed approach provides a scalable, high-performance solution for intelligent task offloading in next-generation 6G edge networks.</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

Hybrid DDQN-Fuzzy SARSA with Experience Replay for URLLC-Aware Task Offloading in 6G Edge Computing Networks

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Arvind Kumar

摘要

The rapid growth of 6G networks has intensified the need for ultra-reliable low-latency communication (URLLC) and intelligent task offloading in multi-edge computing environments. This study proposes an enhanced Hybrid Double Deep Q-Network–Fuzzy SARSA (DDQN–Fuzzy SARSA) model integrating experience replay memory to achieve stable, interpretable, and real-time offloading decisions. The model combines the long-term action-value optimization of DDQN with the adaptive rule-based reasoning of Fuzzy SARSA ( \(\lambda\) ), while a task-failure-aware reward function incorporating Task Failure Rate (TFR) ensures rapid avoidance of congested edge nodes and reinforces QoS guarantees. A comprehensive system model considering MAN/WAN delays, VM utilisation, server coordinates, and heterogeneous IoT workloads was developed and evaluated through extensive MATLAB simulations. Results demonstrate that the proposed framework significantly surpasses standalone DQN, Fuzzy SARSA, PPO, A3C, and PSO baselines, achieving 0.35 ms latency, 99.9999% reliability, 99.67% energy efficiency, and a 40% reduction in task failure rate. The hybrid model also generalises effectively across diverse applications including healthcare, AR/VR, infotainment, and compute-intensive services. Overall, the proposed approach provides a scalable, high-performance solution for intelligent task offloading in next-generation 6G edge networks.