Scholarlink

<p>This study introduces a novel theoretical framework for multi-agent hierarchical reinforcement learning (MAHRL) in cloud resource orchestration. Our key contributions include: (1) developing a sophisticated MAHRL model that efficiently handles large state–action spaces in dynamic cloud environments; (2) establishing a rigorous convergence analysis based on Lyapunov stability theory, providing strong theoretical guarantees for the proposed algorithm; and (3) deriving comprehensive performance bounds, including approximation error bounds and regret bounds, along with a detailed computational complexity analysis. Through a combination of theoretical proofs and extensive numerical simulations using realistic workload traces, we demonstrated the superior performance, stability, and scalability of our MAHRL framework. Our results show improvements in resource utilization (15–20%), energy efficiency (34.4% reduction), and service quality (40% fewer SLA violations).</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

A theoretical framework for multi-agent hierarchical reinforcement learning in cloud resource orchestration with convergence guarantees

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Chunmao Jiang,
  • Longmei Tang,
  • Ruyi Ye,
  • Hao Zhang

摘要

This study introduces a novel theoretical framework for multi-agent hierarchical reinforcement learning (MAHRL) in cloud resource orchestration. Our key contributions include: (1) developing a sophisticated MAHRL model that efficiently handles large state–action spaces in dynamic cloud environments; (2) establishing a rigorous convergence analysis based on Lyapunov stability theory, providing strong theoretical guarantees for the proposed algorithm; and (3) deriving comprehensive performance bounds, including approximation error bounds and regret bounds, along with a detailed computational complexity analysis. Through a combination of theoretical proofs and extensive numerical simulations using realistic workload traces, we demonstrated the superior performance, stability, and scalability of our MAHRL framework. Our results show improvements in resource utilization (15–20%), energy efficiency (34.4% reduction), and service quality (40% fewer SLA violations).