Scholarlink

<p>The galloping movement toward completely renewable power systems means that new control architectures are needed that can cope with increased variability to date whilst maintaining short-lived stability in low-inertia networks. This paper presents a Neural Grid Control System (NGCS) that combines physics-informed graph attention networks with hierarchical multi-agent reinforcement learning to provide enhanced predictive transient stability and achieve real-time topology reconfiguration. The proposed architecture uses a spatiotemporal graph encoder, which is able to capture the dynamic electrical coupling at grid-forming converter-dominated networks and embeds the physical constraints of swing equation within the learning architecture to ensure physical consistency. An intelligent layer with novel distributed intelligence enables autonomous agents at the autonomous substation level to synchronize topology-switching behavior using consensus-based communication schemes, where a centralized meta-controller coordinates system-wide stability goals. The proposed NGCS demonstrates 94.7% transient stability classification accuracy, achieves inference in 12 ms, and provides a 340 computational speedup compared to conventional time-domain simulation, enabling real-time topology reconfiguration in fully inverter-based grids. These results indicate that NGCS is a promising framework for fast transient-stability assessment and coordinated corrective control in converter-dominated power systems.</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

Neural grid control systems with predictive transient stability enhancement enabling 100 percent renewable integration through distributed intelligence and real time topology reconfiguration

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月15日

查看全文
  • Zhiliang Cui,
  • Yuze Liang

摘要

The galloping movement toward completely renewable power systems means that new control architectures are needed that can cope with increased variability to date whilst maintaining short-lived stability in low-inertia networks. This paper presents a Neural Grid Control System (NGCS) that combines physics-informed graph attention networks with hierarchical multi-agent reinforcement learning to provide enhanced predictive transient stability and achieve real-time topology reconfiguration. The proposed architecture uses a spatiotemporal graph encoder, which is able to capture the dynamic electrical coupling at grid-forming converter-dominated networks and embeds the physical constraints of swing equation within the learning architecture to ensure physical consistency. An intelligent layer with novel distributed intelligence enables autonomous agents at the autonomous substation level to synchronize topology-switching behavior using consensus-based communication schemes, where a centralized meta-controller coordinates system-wide stability goals. The proposed NGCS demonstrates 94.7% transient stability classification accuracy, achieves inference in 12 ms, and provides a 340 computational speedup compared to conventional time-domain simulation, enabling real-time topology reconfiguration in fully inverter-based grids. These results indicate that NGCS is a promising framework for fast transient-stability assessment and coordinated corrective control in converter-dominated power systems.