pith. sign in

arxiv: 2507.23094 · v3 · pith:534Q5G4Dnew · submitted 2025-07-30 · 🧮 math.OC

Stability-Constrained AC Optimal Power Flow--A Gaussian Process-Based Approach

Vincenzo Di Vito , Kaarthik Sundar , Ferdinando Fioretto , Deepjyoti Deka This is my paper

Pith reviewed 2026-05-19 02:06 UTC · model grok-4.3

classification 🧮 math.OC
keywords ACOPFGaussian processdynamic stabilitysurrogate modelingpower system optimizationgenerator dynamicsstability constraintsoptimal power flow
0
0 comments X

The pith

A Gaussian process learns an exponential stability surrogate to constrain AC optimal power flow to dynamically safe points.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a data-driven method to add generator dynamics to the standard AC optimal power flow problem. Conventional solutions can be unstable because they ignore how generators respond over time. The approach fits a Gaussian process to learn an exponential function that signals stability from bus voltage values, using data from differential equation simulations. This learned function then becomes a constraint in the optimization, allowing selection of operating points that are both low-cost and stable. Tests on IEEE test systems up to 118 buses show it works with small training sets and gives more reliable results than other data-driven techniques.

Core claim

The central claim is that generator stability can be characterized by an exponential surrogate whose exponent is learned via Gaussian process regression as a function of bus voltage from limited training data generated by solving the governing differential equations. Integrating this surrogate into the ACOPF allows the formulation to identify operating points satisfying both operational safety and dynamic stability criteria through probabilistic assessment.

What carries the argument

An exponential surrogate function for stability, with its exponent modeled by a Gaussian process regression on bus voltage data.

If this is right

  • The dynamics-aware ACOPF yields operating points that satisfy both cost and stability criteria.
  • Probabilistic stability assessment becomes part of the optimization process.
  • The approach captures generator dynamics efficiently using limited training data.
  • It yields more reliable decisions across a range of operating conditions than existing data-driven methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The surrogate model could be updated online as new dynamic data becomes available in operational settings.
  • This technique might apply to other optimization problems in power systems that involve stability, such as security-constrained dispatch.
  • Extending the voltage-based surrogate to include other state variables could improve accuracy for complex dynamics.

Load-bearing premise

The exponential surrogate function learned via GP regression on limited training data from the differential equations accurately characterizes stability across the full range of operating conditions and system sizes tested.

What would settle it

A full nonlinear dynamic simulation of the grid starting from one of the optimized points that would show growing oscillations if the surrogate incorrectly classified it as stable.

Figures

Figures reproduced from arXiv: 2507.23094 by Deepjyoti Deka, Ferdinando Fioretto, Kaarthik Sundar, Vincenzo Di Vito.

Figure 1
Figure 1. Figure 1: In stable operating conditions, the rotor angle [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The exponential surrogate fit effectively captures the [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The time taken in seconds to solve the GP-SC-ACOPF [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
read the original abstract

The Alternating Current Optimal Power Flow (ACOPF) problem is a core task in power system operations, aimed at determining cost-effective generation dispatch while satisfying physical and operational constraints. However, conventional ACOPF formulations rely on steady-state models and neglect generator dynamics, which can result in operating points that are economically optimal but dynamically unstable. This paper proposes a novel, data-driven approach to incorporate generator dynamics into the ACOPF using Gaussian Process (GP) models. Specifically, it introduces an exponential surrogate function to characterize the stability of solutions to the differential equations governing synchronous generator dynamics. The exponent, which indicates whether system trajectories decay (stable) or grow (unstable), is learned as a function of the bus voltage using GP regression. Crucially, the framework enables probabilistic stability assessment to be integrated directly into the optimization process. The resulting dynamics-aware ACOPF formulation identifies operating points that satisfy both operational safety and dynamic stability criteria. Numerical experiments on the IEEE 39-bus, 57-bus, and 118-bus systems demonstrate that, compared with existing data-driven approaches, the proposed method efficiently captures generator dynamics with limited training data, yielding more reliable and robust decisions across a wide range of operating conditions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a data-driven method to incorporate synchronous generator dynamics into the AC Optimal Power Flow (ACOPF) problem. It learns an exponential surrogate function via Gaussian Process (GP) regression that maps bus voltage to a stability exponent (indicating decay or growth of trajectories), then embeds a probabilistic stability constraint into the ACOPF formulation. Numerical results on the IEEE 39-, 57-, and 118-bus systems are reported to show that the resulting operating points satisfy both operational limits and dynamic stability criteria more reliably than prior data-driven baselines, using limited training data.

Significance. If the surrogate proves reliable for optimizer-selected points, the work would offer a practical route to dynamics-aware ACOPF without repeated time-domain simulations inside the optimizer. The explicit use of GP for probabilistic stability assessment and the demonstration across three standard test systems with modest training data are concrete strengths that could influence how stability constraints are handled in real-time market and security-constrained dispatch tools.

major comments (2)
  1. [§3.2] §3.2 (exponential surrogate definition): the stability indicator is learned solely as a function of bus voltage magnitude. Because the underlying differential equations also depend on rotor angles, speeds, and internal voltages, the surrogate can label a voltage profile as stable while the full state trajectory is unstable; the non-convex ACOPF can then exploit any such misclassified region.
  2. [§5] §5 (numerical experiments): the reported improvements in reliability are not accompanied by direct time-domain simulation verification of the final optimized points, nor by error bars or sensitivity analysis on the GP hyperparameters and training-set size. Without these checks it is unclear whether the claimed stability satisfaction holds for the points actually returned by the solver.
minor comments (1)
  1. [§4] The description of how the GP posterior is converted into a deterministic constraint inside the ACOPF could be expanded with an explicit equation reference for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We address each major comment below and indicate the revisions planned for the manuscript.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (exponential surrogate definition): the stability indicator is learned solely as a function of bus voltage magnitude. Because the underlying differential equations also depend on rotor angles, speeds, and internal voltages, the surrogate can label a voltage profile as stable while the full state trajectory is unstable; the non-convex ACOPF can then exploit any such misclassified region.

    Authors: We acknowledge that the surrogate is formulated as a function of bus voltage magnitude alone. In the ACOPF setting the voltage profile is the decision variable, and the power-flow equations implicitly relate this profile to rotor angles and internal voltages at the steady-state operating point from which the dynamics are initialized. Nevertheless, we agree that this modeling choice introduces an approximation whose limitations should be discussed explicitly. We will revise §3.2 to add a paragraph clarifying the underlying assumptions, noting the dependence on other states, and stating the conditions under which the voltage-only surrogate remains reliable for the test systems considered. revision: partial

  2. Referee: [§5] §5 (numerical experiments): the reported improvements in reliability are not accompanied by direct time-domain simulation verification of the final optimized points, nor by error bars or sensitivity analysis on the GP hyperparameters and training-set size. Without these checks it is unclear whether the claimed stability satisfaction holds for the points actually returned by the solver.

    Authors: We agree that explicit verification strengthens the claims. Although internal time-domain checks were performed during development, they were not reported in detail. We will revise §5 to include (i) direct time-domain simulation results confirming stability of the final ACOPF solutions, (ii) error bars on the stability-exponent predictions, and (iii) sensitivity plots with respect to GP length-scale, noise variance, and training-set size. These additions will be presented for all three test systems. revision: yes

Circularity Check

0 steps flagged

No significant circularity; data-driven surrogate is explicit approximation

full rationale

The paper presents an explicitly data-driven method: Gaussian Process regression is used to fit an exponential surrogate for the stability exponent directly from simulation trajectories of the generator differential equations, with the surrogate then inserted as a probabilistic constraint inside the ACOPF. This is a standard modeling approximation rather than a derivation that reduces to its own inputs by construction. No equations are shown to be equivalent to the training data by definition, no load-bearing self-citation chain is invoked, and no uniqueness theorem or ansatz is smuggled in. The numerical results on IEEE test systems constitute external validation of the fitted surrogate's behavior. Concerns about generalization or fidelity of the voltage-only mapping are correctness issues, not circularity.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The approach depends on the validity of the exponential surrogate as a stability proxy and on the GP regression assumptions holding for the generator dynamics data.

free parameters (1)
  • Gaussian Process hyperparameters
    Fitted during regression to map bus voltage to stability exponent.
axioms (1)
  • standard math Gaussian Process regression assumptions (smoothness, kernel choice)
    Invoked when learning the stability function from simulation data.
invented entities (1)
  • exponential surrogate function for stability no independent evidence
    purpose: Maps generator trajectory decay/growth to a scalar exponent usable as optimization constraint
    Introduced to turn differential-equation stability into a learnable function of voltage.

pith-pipeline@v0.9.0 · 5755 in / 1195 out tokens · 55208 ms · 2026-05-19T02:06:29.070768+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages · 3 internal anchors

  1. [1]

    Taha, and Josh A

    Mohammadhafez Bazrafshan, Nikolaos Gatsis, Ahmad F. Taha, and Josh A. Taylor. Augmenting the optimal power flow for stability. In 2016 IEEE 55th Conference on Decision and Control (CDC) , pages 4104–4109, 2016. doi: 10.1109/CDC.2016.7798891

    work page doi:10.1109/cdc.2016.7798891 2016

  2. [2]

    Discretize then optimize

    John T Betts and Stephen L Campbell. Discretize then optimize. Mathematics for industry: challenges and frontiers , pages 140–157, 2005

    work page 2005

  3. [3]

    Bynum, Gabriel A

    Michael L. Bynum, Gabriel A. Hackebeil, William E. Hart, Carl D. Laird, Bethany L. Nicholson, John D. Siirola, Jean-Paul Watson, and David L. Woodruff. Pyomo–optimization modeling in python , vol- ume 67. Springer Science & Business Media, third edition, 2021

    work page 2021

  4. [4]

    Robust deep gaussian process-based probabilistic electrical load forecasting against anomalous events

    Di Cao, Junbo Zhao, Weihao Hu, Yingchen Zhang, Qishu Liao, Zhe Chen, and Frede Blaabjerg. Robust deep gaussian process-based probabilistic electrical load forecasting against anomalous events. IEEE Transactions on Industrial Informatics , 18(2):1142–1153, 2022. doi: 10.1109/TII.2021.3081531

    work page doi:10.1109/tii.2021.3081531 2022

  5. [5]

    State-of-the-art, challenges, and future trends in security constrained optimal power flow

    Florin Capitanescu, JL Martinez Ramos, Patrick Panciatici, Daniel Kirschen, A Marano Marcolini, Ludovic Platbrood, and Louis Wehenkel. State-of-the-art, challenges, and future trends in security constrained optimal power flow. Electric power systems research , 81(8):1731–1741, 2011

    work page 2011

  6. [6]

    Ricky T. Q. Chen. torchdiffeq, 2018. URL https://github.com/rtqichen/ torchdiffeq

    work page 2018

  7. [7]

    Learning to optimize meets neural-ode: Real-time, stability- constrained ac opf, 2024

    Vincenzo Di Vito, Mostafa Mohammadian, Kyri Baker, and Ferdinando Fioretto. Learning to optimize meets neural-ode: Real-time, stability- constrained ac opf, 2024. URL https://arxiv.org/abs/2410.19157

    work page arXiv 2024

  8. [8]

    Gpytorch: Blackbox matrix-matrix gaussian process inference with gpu acceleration

    Jacob R Gardner, Geoff Pleiss, David Bindel, Kilian Q Weinberger, and Andrew Gordon Wilson. Gpytorch: Blackbox matrix-matrix gaussian process inference with gpu acceleration. In Advances in Neural Infor- mation Processing Systems , 2018

    work page 2018

  9. [9]

    Physics- informed variational state-space Gaussian processes

    Oliver Hamelijnck, Arno Solin, and Theodoros Damoulas. Physics- informed variational state-space Gaussian processes. In Advances in Neural Information Processing Systems (NeurIPS) , 2024

    work page 2024

  10. [10]

    Definition and classification of power system stability – revisited & extended

    Nikos Hatziargyriou et al. Definition and classification of power system stability – revisited & extended. IEEE Transactions on Power Systems , 36(4):3271–3281, 2021. doi: 10.1109/TPWRS.2020.3041774

    work page doi:10.1109/tpwrs.2020.3041774 2021

  11. [11]

    Learning unknown ODE models with Gaussian processes

    Markus Heinonen, Cagatay Yildiz, Henrik Mannerström, Jukka In- tosalmi, and Harri Lähdesmäki. Learning unknown ode models with gaussian processes, 2018. URL https://arxiv.org/abs/1803.04303

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  12. [12]

    Robust multi-period opf with storage and renewables

    Rabih A Jabr, Sami Karaki, and Joe Akl Korbane. Robust multi-period opf with storage and renewables. IEEE Transactions on Power Systems , 30(5):2790–2799, 2014

    work page 2014

  13. [13]

    Hassan K. Khalil. Nonlinear Systems . Prentice Hall, 3rd edition, 2002

    work page 2002

  14. [14]

    Modelling and control of dynamic systems using Gaussian process models

    Juš Kocijan. Modelling and control of dynamic systems using Gaussian process models. Springer, 2016

    work page 2016

  15. [15]

    P. Kundur. Power System Stability and Control . McGraw-Hill, 1994

    work page 1994

  16. [16]

    Explicit data-driven small-signal stability constrained 12 optimal power flow

    Juelin Liu, Zhifang Yang, Junbo Zhao, Juan Yu, Bendong Tan, and Wenyuan Li. Explicit data-driven small-signal stability constrained 12 optimal power flow. IEEE Transactions on Power Systems , 37(5):3726– 3737, 2021

    work page 2021

  17. [17]

    Power system dynamics: stability and control

    Jan Machowski, Zbigniew Lubosny, Janusz W Bialek, and James R Bumby. Power system dynamics: stability and control . John Wiley & Sons, 2020

    work page 2020

  18. [18]

    Markwardt

    Craig B. Markwardt. Non-linear least squares fitting in idl with mpfit,

    work page

  19. [19]

    URL https://arxiv.org/abs/0902.2850

    work page internal anchor Pith review Pith/arXiv arXiv

  20. [20]

    Data-driven stochastic ac-opf using gaussian process regression

    Mile Mitrovic, Aleksandr Lukashevich, Petr V orobev, Vladimir Terzija, Semen Budennyy, Yury Maximov, and Deepjyoti Deka. Data-driven stochastic ac-opf using gaussian process regression. International Journal of Electrical Power & Energy Systems , 152:109249, 2023

    work page 2023

  21. [21]

    A survey of relaxations and approximations of the power flow equations

    Daniel K Molzahn, Ian A Hiskens, et al. A survey of relaxations and approximations of the power flow equations. F oundations and Trends® in Electric Energy Systems , 4(1-2):1–221, 2019

    work page 2019

  22. [22]

    Security-constrained optimal power flow with post-contingency corrective rescheduling

    A Monticelli, MVF Pereira, and S Granville. Security-constrained optimal power flow with post-contingency corrective rescheduling. IEEE Transactions on Power Systems , 2(1):175–180, 2007

    work page 2007

  23. [23]

    Parikshit Pareek and Hung D. Nguyen. A framework for analytical power flow solution using gaussian process learning. IEEE Transactions on Sustainable Energy , 13(1):452–463, 2022. doi: 10.1109/TSTE.2021. 3116544

    work page doi:10.1109/tste.2021 2022

  24. [24]

    Fast risk assess- ment in power grids through novel gaussian process and active learning

    Parikshit Pareek, Deepjyoti Deka, and Sidhant Misra. Fast risk assess- ment in power grids through novel gaussian process and active learning. arXiv preprint arXiv:2308.07867 , 2023

    work page arXiv 2023

  25. [25]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Brad- bury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. Pytorch: An imperative style, high-perfor...

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  26. [26]

    Machine learning of linear differential equations using gaussian processes

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Machine learning of linear differential equations using gaussian processes. Jour- nal of Computational Physics , 348:683–693, November 2017. ISSN 0021-9991. doi: 10.1016/j.jcp.2017.07.050. URL http://dx.doi.org/10. 1016/j.jcp.2017.07.050

    work page doi:10.1016/j.jcp.2017.07.050 2017

  27. [27]

    Carl Edward Rasmussen and Christopher K. I. Williams. Gaussian Processes for Machine Learning . MIT Press, 2006. URL http://www. gaussianprocess.org/gpml/

    work page 2006

  28. [28]

    Sauer and M

    Peter W. Sauer and M. A. Pai. Power System Dynamics and Stability . Prentice Hall, Upper Saddle River, N.J., 1998

    work page 1998

  29. [29]

    Analytic input convex neural networks-based model predictive control for power system transient stability enhancement

    Tong Su, Junbo Zhao, Xiao Chen, and Xiaodong Liu. Analytic input convex neural networks-based model predictive control for power system transient stability enhancement. In Proc. IEEE Power & Energy Society General Meeting (PESGM) , 2023. doi: 10.1109/PESGM52003.2023. 10252514

    work page doi:10.1109/pesgm52003.2023 2023

  30. [30]

    Safe reinforcement learning-based transient stability control for islanded microgrids with topology reconfiguration

    Tong Su, Junbo Zhao, Yiyun Yao, Alaa Selim, and Fei Ding. Safe reinforcement learning-based transient stability control for islanded microgrids with topology reconfiguration. IEEE Transactions on Smart Grid, 16(4):3432–3444, 2025. doi: 10.1109/TSG.2025.3569696

    work page doi:10.1109/tsg.2025.3569696 2025

  31. [31]

    Álvarez, and Neil D

    Simo Särkkä, Mauricio A. Álvarez, and Neil D. Lawrence. Gaussian process latent force models for learning and stochastic control of physical systems. IEEE Transactions on Automatic Control, 64(7):2953– 2960, 2019. doi: 10.1109/TAC.2018.2874749

    work page doi:10.1109/tac.2018.2874749 2019

  32. [32]

    Gaussian processes in power systems: Techniques, applications, and future works,

    Bendong Tan, Tong Su, Yu Weng, Ketian Ye, Parikshit Pareek, Petr V orobev, Hung Nguyen, Junbo Zhao, and Deepjyoti Deka. Gaussian processes in power systems: Techniques, applications, and future works,

    work page

  33. [33]

    URL https://arxiv.org/abs/2505.15950

    work page arXiv

  34. [34]

    Scalable risk assessment of rare events in power systems with uncertain wind generation and loads

    Bendong Tan, Junbo Zhao, and Yousu Chen. Scalable risk assessment of rare events in power systems with uncertain wind generation and loads. IEEE Transactions on Power Systems , 40(2):1374–1388, 2025. doi: 10.1109/TPWRS.2024.3435490

    work page doi:10.1109/tpwrs.2024.3435490 2025

  35. [35]

    E., et al

    Pauli Virtanen, Ralf Gommers, Travis E. Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, Stéfan J. van der Walt, Matthew Brett, Joshua Wilson, K. Jarrod Millman, Nikolay Mayorov, Andrew R. J. Nelson, Eric Jones, Robert Kern, Eric Larson, C J Carey, ˙Ilhan Polat, Yu Feng, Eric W. M...

    work page doi:10.1038/s41592-019-0686-2 2020

  36. [36]

    Andreas Wächter and Lorenz T. Biegler. On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical Programming, 106:25–57, 2006

    work page 2006

  37. [37]

    A robust approach to warped gaussian process-constrained optimization, 2020

    Johannes Wiebe, Inês Cecílio, Jonathan Dunlop, and Ruth Misener. A robust approach to warped gaussian process-constrained optimization, 2020

    work page 2020

  38. [38]

    Distribution- ally Robust Chance-Constrained AC-OPF for Integrating Wind Energy Through Multi-Terminal VSC-HVDC

    Li Yao, Xiuli Wang, Yujun Li, Chao Duan, and Xiong Wu. Distribution- ally Robust Chance-Constrained AC-OPF for Integrating Wind Energy Through Multi-Terminal VSC-HVDC. IEEE Transactions on Sustain- able Energy, 11(3):1414–1426, 2020. doi: 10.1109/TSTE.2019.2927135

    work page doi:10.1109/tste.2019.2927135 2020

  39. [39]

    Physics- informed sparse gaussian process for probabilistic stability analysis of large-scale power system with dynamic pvs and loads

    Ketian Ye, Junbo Zhao, Nan Duan, and Yingchen Zhang. Physics- informed sparse gaussian process for probabilistic stability analysis of large-scale power system with dynamic pvs and loads. IEEE Transactions on Power Systems , 38(3):2868–2879, 2023. doi: 10.1109/ TPWRS.2022.3188182

    work page arXiv 2023

  40. [40]

    A high computationally efficient parallel partial gaussian process for large- scale power system probabilistic transient stability assessment

    Ketian Ye, Junbo Zhao, Hanmo Li, and Mengyang Gu. A high computationally efficient parallel partial gaussian process for large- scale power system probabilistic transient stability assessment. IEEE Transactions on Power Systems , 39(2):4650–4660, 2024. doi: 10.1109/ TPWRS.2023.3320130

    work page arXiv 2024

  41. [41]

    Matpower: Steady-state operations, planning, and analysis tools for power systems research and education

    Ray Daniel Zimmerman, Carlos Edmundo Murillo-Sánchez, and Robert John Thomas. Matpower: Steady-state operations, planning, and analysis tools for power systems research and education. IEEE Transactions on power systems , 26(1):12–19, 2010

    work page 2010