Scholarlink

<p>Resilience of unmanned weapon system-of-systems (UWSoS) is critical for preserving sustained operational capability amidst adversarial disruptions and cascading subsystem degradations. However, prevailing resilience enhancement methodologies frequently overlook the semantic richness embedded within meta-paths spanning the heterogeneous network, thereby constraining their adaptability and robustness. This study introduces meta-path-driven reinforcement learning(MPD-RL), an innovative framework that synergizes meta-path-guided node embedding with deep reinforcement learning to surmount this limitation. Within MPD-RL, a topology-informed embedding mechanism utilizes meta-path-based random walks and structurally derived matrices to encapsulate high-order relational semantics intrinsic to the UWSoS architecture. Furthermore, an auxiliary loss function is introduced to embed meta-path knowledge into the reinforcement learning agent’s optimization process, thereby steering policy development toward resilience-oriented decision making. By incorporating meta-path semantics into both the agent’s perceptual state and objective function, MPD-RL facilitates proactive identification and fortification of critical communication pathways, enabling sustained network performance under adverse conditions. Extensive simulation results substantiate that MPD-RL markedly surpasses contemporary baseline techniques in preserving network connectivity and functional integrity across diverse failure and attack scenarios.</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

MPD-RL: Meta-path-driven reinforcement learning for enhancing resilience in unmanned weapon system-of-systems

  • 期刊论文

  • 发表日期：

    2025年6月2日

查看全文
  • Zaikun Han,
  • Siwen Wei,
  • Dingrui Xue,
  • Yuqing Lin,
  • Ke Li,
  • Jiancheng Liu,
  • Xingye Han,
  • Gang Hou,
  • Junxiong Ye,
  • Kexin Wang

摘要

Resilience of unmanned weapon system-of-systems (UWSoS) is critical for preserving sustained operational capability amidst adversarial disruptions and cascading subsystem degradations. However, prevailing resilience enhancement methodologies frequently overlook the semantic richness embedded within meta-paths spanning the heterogeneous network, thereby constraining their adaptability and robustness. This study introduces meta-path-driven reinforcement learning(MPD-RL), an innovative framework that synergizes meta-path-guided node embedding with deep reinforcement learning to surmount this limitation. Within MPD-RL, a topology-informed embedding mechanism utilizes meta-path-based random walks and structurally derived matrices to encapsulate high-order relational semantics intrinsic to the UWSoS architecture. Furthermore, an auxiliary loss function is introduced to embed meta-path knowledge into the reinforcement learning agent’s optimization process, thereby steering policy development toward resilience-oriented decision making. By incorporating meta-path semantics into both the agent’s perceptual state and objective function, MPD-RL facilitates proactive identification and fortification of critical communication pathways, enabling sustained network performance under adverse conditions. Extensive simulation results substantiate that MPD-RL markedly surpasses contemporary baseline techniques in preserving network connectivity and functional integrity across diverse failure and attack scenarios.