Mohammadi, Sina, Hoang, Quang Manh, Wada, Marcus Chen I, Bui, Van-Hai, Su, Wencong · Nov 2, 2026
arXiv: Grid-Mode-Aware AI Data Center Power Smoothing via Bayesian-Optimized Hybrid Energy Storage System
Markdown · JSON · Original source
The paper’s authors propose a real- control framework for smoothing AI data center power demand, which they say can fluctuate sharply during intensive workloads and synchronized training cycles. Their approach coordinates a Battery Energy Storage System (BESS) and a supercapacitor (SC) as a Hybrid Energy Storage System (HESS), using a frequency-decoupling and active modal-damping policy to reduce oscillatory load components and smooth overall fluctuations. Rather than use online receding-horizon optimization, the framework uses offline Bayesian Optimization to tune control parameters against a multi-objective cost function balancing grid compliance penalties and storage degradation. In numerical simulations, the authors report reshaping raw demand fluctuations of 75–160 to largely fit within a 10 MW target band, with brief transient excursions, and reducing a 20 MW oscillatory component to below 1 MW. They present these results as a way to suppress resonant excitation and maintain grid stability. The evidence described in the abstract is simulation-based; it does not establish performance in operating data centers or power grids.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
eess.SY · Article
Source details
Original source date: Nov 2, 2026 (2026-11-02).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T07:03:44.419Z. First observed: 2026-11-03T08:51:33.513Z. Retrieved: 2026-11-03T14:33:13.425Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The abstract reports numerical simulations; it does not establish that the proposed method has been tested in operating data centers or power grids. The evidence provided here does not describe the simulation conditions or the behavior of the proposed framework outside the reported scenarios.
Open question
How would the reported smoothing and oscillation reduction hold up in operating data centers and power grids?
Source-stated
The authors propose a real-time control framework that coordinates a Battery Energy Storage System and a supercapacitor to smooth AI data center power fluctuations.
This paper proposes a real-time, forecasting-independent control framework utilizing a Hybrid Energy Storage System (HESS) that coordinates a Battery Energy Storage System (BESS) and a supercapacitor (SC).
Source-stated
The framework uses offline Bayesian Optimization to tune real-time control parameters, with a cost function balancing grid compliance penalties and storage degradation.
this work introduces an offline Bayesian Optimization (BO) approach to optimally tune real-time control parameters, minimizing a multi-objective cost function that balances grid compliance penalties and storage degradation.
Source-stated
In numerical simulations, the authors report reshaping 75–160 MW raw demand fluctuations to largely fit within a 10 MW target band, with brief transient excursions.
it reshapes extreme 75-160 MW raw demand fluctuations to largely fit within a tight 10 MW target band, with only brief transient excursions
Source-stated
In numerical simulations, the authors report attenuating a 20 MW oscillatory component to below 1 MW.
while attenuating the critical 20 MW oscillatory component below 1 MW to suppress resonant excitation.
Source and contributor context
Mohammadi, Sina, Hoang, Quang Manh, Wada, Marcus Chen I, Bui, Van-Hai, Su, Wencong · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Megawatt
A unit of power equal to a million watts.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: d0a0f47f-6ea0-430c-8912-83367af13196.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T07:03:44.419Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 81ab3bd7e320a102014e2c70b6af099159fa241164f1ea172d1f5218928d8d9e. Source text hash: bbc71d46c5cc84956ef1649ec9c4c953c7c5a01842c1ba26bc328b471ac83499.
Revision and source-check history
Version 1 · 2026-11-10T07:03:44.419Z
arXiv: Grid-Mode-Aware AI Data Center Power Smoothing via Bayesian-Optimized Hybrid Energy Storage System
The paper’s authors propose a real- control framework for smoothing AI data center power demand, which they say can fluctuate sharply during intensive workloads and synchronized training cycles. Their approach coordinates a Battery Energy Storage System (BESS) and a supercapacitor (SC) as a Hybrid Energy Storage System (HESS), using a frequency-decoupling and active modal-damping policy to reduce oscillatory load components and smooth overall fluctuations. Rather than use online receding-horizon optimization, the framework uses offline Bayesian Optimization to tune control parameters against a multi-objective cost function balancing grid compliance penalties and storage degradation. In numerical simulations, the authors report reshaping raw demand fluctuations of 75–160 to largely fit within a 10 MW target band, with brief transient excursions, and reducing a 20 MW oscillatory component to below 1 MW. They present these results as a way to suppress resonant excitation and maintain grid stability. The evidence described in the abstract is simulation-based; it does not establish performance in operating data centers or power grids.
Original source
eess.SY · Article
Source details
Original source date: Nov 2, 2026 (2026-11-02).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T07:03:44.419Z. First observed: 2026-11-03T08:51:33.513Z. Retrieved: 2026-11-03T14:33:13.425Z.
Limits
The abstract reports numerical simulations; it does not establish that the proposed method has been tested in operating data centers or power grids. The evidence provided here does not describe the simulation conditions or the behavior of the proposed framework outside the reported scenarios.
Open question
How would the reported smoothing and oscillation reduction hold up in operating data centers and power grids?
Source-stated
The authors propose a real-time control framework that coordinates a Battery Energy Storage System and a supercapacitor to smooth AI data center power fluctuations.
This paper proposes a real-time, forecasting-independent control framework utilizing a Hybrid Energy Storage System (HESS) that coordinates a Battery Energy Storage System (BESS) and a supercapacitor (SC).
Source-stated
The framework uses offline Bayesian Optimization to tune real-time control parameters, with a cost function balancing grid compliance penalties and storage degradation.
this work introduces an offline Bayesian Optimization (BO) approach to optimally tune real-time control parameters, minimizing a multi-objective cost function that balances grid compliance penalties and storage degradation.
Source-stated
In numerical simulations, the authors report reshaping 75–160 MW raw demand fluctuations to largely fit within a 10 MW target band, with brief transient excursions.
it reshapes extreme 75-160 MW raw demand fluctuations to largely fit within a tight 10 MW target band, with only brief transient excursions
Source-stated
In numerical simulations, the authors report attenuating a 20 MW oscillatory component to below 1 MW.
while attenuating the critical 20 MW oscillatory component below 1 MW to suppress resonant excitation.
Source and contributor context
Mohammadi, Sina, Hoang, Quang Manh, Wada, Marcus Chen I, Bui, Van-Hai, Su, Wencong · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Megawatt
A unit of power equal to a million watts.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: d0a0f47f-6ea0-430c-8912-83367af13196.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T07:03:44.419Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 81ab3bd7e320a102014e2c70b6af099159fa241164f1ea172d1f5218928d8d9e. Source text hash: bbc71d46c5cc84956ef1649ec9c4c953c7c5a01842c1ba26bc328b471ac83499.
uncertain · 2026-11-10T07:03:44.419Z · anthropic/claude-opus-5.5
Publication: d0a0f47f-6ea0-430c-8912-83367af13196
attribution — pass: The reading consistently attributes the proposal and results to the paper's authors on arXiv. Simulation outcomes are framed as what the authors report, not as established fact.
Authors:Sina Mohammadi, Quang Manh Hoang, Marcus Chen I Wada, Van-Hai Bui, Wencong Su
summary — uncertain: The draft is faithful, but it omits the source's core stake: volatile, synchronized GPU training loads can excite grid natural frequencies and threaten stability. That loss makes 'oscillatory load components' and 'resonant excitation' hard for a non-specialist to follow. Central terms are also left unexplained: frequency-decoupling, active modal-damping, receding-horizon optimization and Bayesian Optimization. These need tooltips. The draft drops 'forecasting-independent' too. Smallest repair: restore the grid-resonance risk sentence, with attribution, and add tooltips for these terms.
which can excite grid natural frequencies and threaten system stability
date — pass: The displayed original publication date of 2026-11-02 matches the v1 submission date on arXiv.
[Submitted on 2 Nov 2026]
support — pass: Each claim matches the abstract. The 75–160 MW figure, the 10 MW band, the brief excursions and the 20 MW to below 1 MW result are all supported. The limitation that the evidence is simulation-only is grounded and stated correctly. The question follows from that gap without adding unsupported assumptions.
Numerical simulations demonstrate that the proposed method ensures grid stability and strong target envelope adherence.
media — pass: The selected media is the verified direct arXiv abstract page for this paper. That makes it the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceA unit of power equal to a million watts.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceA unit of power equal to a million watts.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv