Scholarlink

<p>The decarbonization of electric power systems is displacing synchronous generation with inverter-based renewable resources, eroding the inertia and power-system-stabilizer-derived damping that historically suppressed inter-area oscillations. In parallel, hyperscale artificial-intelligence data centers have emerged as large, continuously operated loads whose power draw is controllable on the millisecond timescale through Dynamic Voltage and Frequency Scaling (DVFS). This paper evaluates whether these compute loads can act as wide-area damping actuators. The analysis uses a DAE-consistent bus-input matrix, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\textbf{B}_b=-\textbf{T}^{-1}\textbf{f}_y\textbf{g}_y^{-1}\textbf{g}_{u,b}\)</EquationSource> </InlineEquation>, and validates the construction against the ANDES reduced state matrix and a finite-difference input check. On the Kundur two-area benchmark the DAE-consistent residue ranking preserves the main siting result: Bus&#xa0;10 is the highest-ranked non-generator bus for the 0.647&#xa0;Hz inter-area mode. A classical washout-plus-lead-lag controller raises damping from 3.43% to approximately 6% using the same DAE input. Under this model, modern 10–50&#xa0;ms battery and DVFS actuators require comparable peak modulation for the same small-signal damping target. The data-center wide-area controller still improves on a closest same-system data-center-frequency-droop baseline: at comparable peak MW the droop baseline reaches 4.76% damping, and reaching 6% requires approximately twice the peak modulation. PMU-delay robustness is also improved, with the damping crossing the open-loop value at 443&#xa0;ms for the 10&#xa0;ms DVFS actuator versus 374&#xa0;ms for the 50&#xa0;ms BESS case in the Padé-delay model. Saturation stress tests show that ordinary 2&#xa0;MW headroom does not bind for modal, fault-cleared-equivalent, and line-outage-equivalent disturbances, while generator-trip-equivalent stress can bind the actuator and must be treated as an operating constraint. IEEE 39-bus scale-up confirms the DAE-based ranking procedure and highlights a practical phase-feasibility issue: the best residue buses may require more phase compensation than a single lead-lag stage can supply.</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

Hyperscale Data Centers as Wide-Area Damping Actuators for Low-Inertia Power Systems

  • 期刊论文

  • 发表日期：

    2026年6月27日

查看全文
  • Mohamed Shamseldein

摘要

The decarbonization of electric power systems is displacing synchronous generation with inverter-based renewable resources, eroding the inertia and power-system-stabilizer-derived damping that historically suppressed inter-area oscillations. In parallel, hyperscale artificial-intelligence data centers have emerged as large, continuously operated loads whose power draw is controllable on the millisecond timescale through Dynamic Voltage and Frequency Scaling (DVFS). This paper evaluates whether these compute loads can act as wide-area damping actuators. The analysis uses a DAE-consistent bus-input matrix, \(\textbf{B}_b=-\textbf{T}^{-1}\textbf{f}_y\textbf{g}_y^{-1}\textbf{g}_{u,b}\) , and validates the construction against the ANDES reduced state matrix and a finite-difference input check. On the Kundur two-area benchmark the DAE-consistent residue ranking preserves the main siting result: Bus 10 is the highest-ranked non-generator bus for the 0.647 Hz inter-area mode. A classical washout-plus-lead-lag controller raises damping from 3.43% to approximately 6% using the same DAE input. Under this model, modern 10–50 ms battery and DVFS actuators require comparable peak modulation for the same small-signal damping target. The data-center wide-area controller still improves on a closest same-system data-center-frequency-droop baseline: at comparable peak MW the droop baseline reaches 4.76% damping, and reaching 6% requires approximately twice the peak modulation. PMU-delay robustness is also improved, with the damping crossing the open-loop value at 443 ms for the 10 ms DVFS actuator versus 374 ms for the 50 ms BESS case in the Padé-delay model. Saturation stress tests show that ordinary 2 MW headroom does not bind for modal, fault-cleared-equivalent, and line-outage-equivalent disturbances, while generator-trip-equivalent stress can bind the actuator and must be treated as an operating constraint. IEEE 39-bus scale-up confirms the DAE-based ranking procedure and highlights a practical phase-feasibility issue: the best residue buses may require more phase compensation than a single lead-lag stage can supply.