Scholarlink

<p>Fragmented power systems remain highly vulnerable to frequency instability following unplanned disturbances. These events often arise when generation is geographically concentrated while loads are dispersed. Within this configuration, sudden system splitting leaves areas with either overgeneration or severe deficits. Distributed renewable energy (DRE) could address this imbalance by locating generation closer to demand; yet, their inverter-based interface introduces low inertia that risks worsening instability. This study proposes AI-controlled inverters, particularly using deep reinforcement learning (DRL) algorithm, as localized, fast-acting stabilizers capable of turning low-inertia DRE into frequency-supporting assets. Using the 2024 Sumatra blackout as a real-world testbed, a 600&#xa0;MW central generation trip was simulated, with its capacity redistributed across four DRE deployment schemes (1 × 600&#xa0;MW, 2 × 300&#xa0;MW, 4 × 150&#xa0;MW, 6 × 100&#xa0;MW). Results show that DRL-enabled inverters consistently improved dynamic stability, where frequency overshoot at the impacted substation was reduced from 4.43&#xa0;Hz (base case) to 0.06&#xa0;Hz, while settling time improved from 37.13 to 12.19s. Unexpectedly, the results also revealed that more distributed DREs does not always equate to better performance, highlighting the importance of optimal placement and environment-specific DRL agent live training. These findings validate the role of DRL-enabled DRE as a fast-response stabilizing mechanism in fragmented grids, while opening a new research pathway on adaptive, optimally placed DRL control for real-world deployment.</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

Turning low-inertia inverters into grid stabilizers: a DRL-based control framework for fragmented power systems

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Muhammad Abu Bakar Sidik,
  • Muhammad Irfan Jambak,
  • Melia Sari,
  • Noer Fadzri Perdana Dinata,
  • Makbul Anwari Muhammad Ramli,
  • Adios Purnama

摘要

Fragmented power systems remain highly vulnerable to frequency instability following unplanned disturbances. These events often arise when generation is geographically concentrated while loads are dispersed. Within this configuration, sudden system splitting leaves areas with either overgeneration or severe deficits. Distributed renewable energy (DRE) could address this imbalance by locating generation closer to demand; yet, their inverter-based interface introduces low inertia that risks worsening instability. This study proposes AI-controlled inverters, particularly using deep reinforcement learning (DRL) algorithm, as localized, fast-acting stabilizers capable of turning low-inertia DRE into frequency-supporting assets. Using the 2024 Sumatra blackout as a real-world testbed, a 600 MW central generation trip was simulated, with its capacity redistributed across four DRE deployment schemes (1 × 600 MW, 2 × 300 MW, 4 × 150 MW, 6 × 100 MW). Results show that DRL-enabled inverters consistently improved dynamic stability, where frequency overshoot at the impacted substation was reduced from 4.43 Hz (base case) to 0.06 Hz, while settling time improved from 37.13 to 12.19s. Unexpectedly, the results also revealed that more distributed DREs does not always equate to better performance, highlighting the importance of optimal placement and environment-specific DRL agent live training. These findings validate the role of DRL-enabled DRE as a fast-response stabilizing mechanism in fragmented grids, while opening a new research pathway on adaptive, optimally placed DRL control for real-world deployment.