pith. sign in

arxiv: 2510.22324 · v2 · submitted 2025-10-25 · 📡 eess.SY · cs.SY

Model-Free Power System Stability Enhancement with Dissipativity-Based Neural Control

Yifei Wang , Han Wang , Kehao Zhuang , Keith Moffat , Florian D\"orfler This is my paper

Pith reviewed 2026-05-18 04:16 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords model-free controldissipativityneural networkspower system stabilityvirtual synchronous generatorstransient stabilitydata-driven control
0
0 comments X

The pith

Neural networks trained on input-state data learn dissipativity matrices that enable model-free stabilizing controllers for virtual synchronous generator power systems.

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

The paper tries to establish a model-free method for enhancing transient stability in power systems with virtual synchronous generators. It does this by training neural networks on input-state data to learn dissipativity-characterizing matrices that can be used to design stabilizing controllers. This avoids the need for an accurate model of the grid dynamics and the difficulties of classical Lyapunov or passivity approaches. Incorporating cost function shaping helps meet user-specified performance objectives. The approach is validated through numerical results on a modified Kundur two-area system with all virtual synchronous generators.

Core claim

By training neural networks on input-state data to approximate dissipativity-characterizing matrices, stabilizing controllers can be obtained for nonlinear power systems without an explicit model, improving transient stability for grid-connected virtual synchronous generators.

What carries the argument

Neural network approximation of dissipativity-characterizing matrices that define storage functions for certifying closed-loop stability via dissipativity.

If this is right

  • Stabilizing controllers can be designed without an accurate model of the grid dynamics.
  • The approach addresses cases where finding storage functions for large grids is intractable with classical methods.
  • Cost function shaping aligns controller performance with user-specified objectives.
  • Numerical tests confirm effectiveness on a modified all-VSG Kundur two-area power system.

Where Pith is reading between the lines

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

  • The data-driven learning could support online adaptation if new trajectories are collected during operation.
  • Similar neural approximation of dissipativity properties might apply to other large nonlinear networked systems.
  • Robust generalization of the learned matrices would reduce dependence on detailed system models in practice.

Load-bearing premise

That neural networks trained on finite input-state trajectories can produce dissipativity matrices whose associated storage functions actually certify closed-loop stability for the real nonlinear power system under all relevant disturbances.

What would settle it

Observing loss of stability or failure of the dissipativity inequality under a disturbance outside the training trajectories would falsify the central claim.

Figures

Figures reproduced from arXiv: 2510.22324 by Florian D\"orfler, Han Wang, Kehao Zhuang, Keith Moffat, Yifei Wang.

Figure 1
Figure 1. Figure 1: The architecture of three types of matrix NNs. For simplicity, we only draw one MLP schematic. Different matrix NNs do not share the same MLP. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An overview of our algorithm, which generate a stabilizing control after training NNs that can characterize dissipativity. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic of the SCIB system. As seen in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The effect of proposed control in the SCIB system. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: The effect of proposed control in the two-area system. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: The modified Kundur two-area system with VSGs. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
read the original abstract

The integration of converter-interfaced generation introduces new transient stability challenges to modern power systems. Classical Lyapunov- and scalable passivity-based approaches typically rely on restrictive assumptions, and finding storage functions for large grids is generally considered intractable. Furthermore, most methods require an accurate grid dynamics model. To address these challenges, we propose a model-free, nonlinear, and dissipativity-based controller which, when applied to grid-connected virtual synchronous generators (VSGs), enhances power system transient stability. Using input-state data, we train neural networks to learn dissipativity-characterizing matrices that yield stabilizing controllers. Furthermore, we incorporate cost function shaping to improve the performance with respect to the user-specified objectives. Numerical results on a modified, all-VSG Kundur two-area power system validate the effectiveness of the proposed approach.

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 model-free nonlinear dissipativity-based neural controller for enhancing transient stability in power systems with virtual synchronous generators. Neural networks are trained on input-state data to learn dissipativity-characterizing matrices that yield stabilizing controllers, with cost function shaping added to align with user objectives. Numerical validation is shown on a modified all-VSG Kundur two-area system.

Significance. If substantiated, the work would advance model-free stability methods for grids with high converter-interfaced generation by avoiding accurate models and restrictive assumptions common in Lyapunov or passivity approaches. The neural learning of dissipativity matrices offers a scalable route to controller design where classical storage-function search is intractable.

major comments (2)
  1. [Numerical Results] §4 (Numerical Results): the reported validation demonstrates empirical stabilization on the Kundur system but supplies no information on training data volume, excitation signal design, or post-training verification that the learned matrices satisfy the dissipativity inequalities for the closed-loop nonlinear dynamics under disturbances outside the training set.
  2. [Dissipativity Learning and Controller Synthesis] §3 (Dissipativity Learning and Controller Synthesis): the central stability claim rests on the storage function associated with the fitted matrices satisfying the dissipativity inequality along every closed-loop trajectory of the true system; finite-trajectory training enforces the condition only approximately at sampled points, with no additional constraint, bound, or out-of-distribution check provided to guarantee the property holds globally.
minor comments (1)
  1. [Introduction] Notation for the learned matrices and storage function could be introduced earlier with an explicit link to the dissipativity supply rate to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment point by point below, indicating where revisions will be made.

read point-by-point responses

  1. Referee: [Numerical Results] §4 (Numerical Results): the reported validation demonstrates empirical stabilization on the Kundur system but supplies no information on training data volume, excitation signal design, or post-training verification that the learned matrices satisfy the dissipativity inequalities for the closed-loop nonlinear dynamics under disturbances outside the training set.

    Authors: We agree that the numerical results section would benefit from greater detail on the data collection and verification process to improve reproducibility. In the revised manuscript, we will expand §4 to report the volume of training data (number of trajectories and samples), describe the excitation signal design (including the specific disturbances and reference changes applied), and add post-training verification results on out-of-distribution disturbances to confirm that the learned dissipativity matrices continue to satisfy the required inequalities for the closed-loop system. revision: yes

  2. Referee: [Dissipativity Learning and Controller Synthesis] §3 (Dissipativity Learning and Controller Synthesis): the central stability claim rests on the storage function associated with the fitted matrices satisfying the dissipativity inequality along every closed-loop trajectory of the true system; finite-trajectory training enforces the condition only approximately at sampled points, with no additional constraint, bound, or out-of-distribution check provided to guarantee the property holds globally.

    Authors: The referee correctly notes that our approach is data-driven and that finite sampled trajectories provide only approximate enforcement of the dissipativity condition at those points. The manuscript does not claim or provide a rigorous global guarantee for all possible closed-loop trajectories, as this would require either a system model or additional theoretical machinery that conflicts with the model-free objective. In the revision we will add an explicit discussion of this limitation in §3, along with empirical out-of-distribution checks in §4 to support practical generalization. A formal bound or constraint guaranteeing the property everywhere is not feasible within the current model-free framework without introducing restrictive assumptions we deliberately avoid. revision: partial

Circularity Check

0 steps flagged

Data-driven dissipativity learning for VSG control is self-contained with empirical validation

full rationale

The paper presents a model-free method that trains neural networks on finite input-state trajectories to obtain dissipativity-characterizing matrices, then applies standard dissipativity theory to synthesize a controller. The central claim is supported by numerical experiments on a modified Kundur two-area system rather than by a closed-form derivation that reduces to the training data by construction. No equation is shown to be identical to its inputs, no fitted parameter is relabeled as an independent prediction, and no load-bearing step relies on a self-citation chain whose content is unverified. The approach is therefore self-contained against external benchmarks (the simulated closed-loop trajectories) and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The approach rests on standard dissipativity theory and the assumption that finite data trajectories suffice to learn globally valid storage functions; no new physical entities are postulated.

free parameters (1)
  • neural-network weights and biases
    Fitted during training to produce the dissipativity-characterizing matrices from collected input-state data.
axioms (1)
  • domain assumption Dissipativity theory supplies Lyapunov-like certificates for nonlinear stability when suitable storage functions exist.
    Invoked to translate learned matrices into stabilizing feedback.

pith-pipeline@v0.9.0 · 5674 in / 1261 out tokens · 37519 ms · 2026-05-18T04:16:36.231278+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

29 extracted references · 29 canonical work pages · 2 internal anchors

  1. [1]

    Foundations and challenges of low-inertia systems,

    F. Milano, F. D ¨orfler, G. Hug, D. J. Hill, and G. Verbiˇc, “Foundations and challenges of low-inertia systems,” in2018 power systems computation conference (PSCC). IEEE, 2018, pp. 1–25

    work page 2018

  2. [2]

    Modelling, implementation, and assessment of virtual synchronous generator in power systems,

    M. Chen, D. Zhou, and F. Blaabjerg, “Modelling, implementation, and assessment of virtual synchronous generator in power systems,”Journal of Modern Power Systems and Clean Energy, vol. 8, no. 3, pp. 399–411, 2020

    work page 2020

  3. [3]

    Analysis of subsynchronous torsional interaction with power electronic converters,

    L. Harnefors, “Analysis of subsynchronous torsional interaction with power electronic converters,”IEEE Transactions on power systems, vol. 22, no. 1, pp. 305–313, 2007

    work page 2007

  4. [4]

    The role of power electronics in future low inertia power systems,

    J. Fang, Y . Tang, H. Li, and F. Blaabjerg, “The role of power electronics in future low inertia power systems,” in2018 IEEE International Power Electronics and Application Conference and Exposition (PEAC). IEEE, 2018, pp. 1–6

    work page 2018

  5. [5]

    A grid- compatible virtual oscillator controller: Analysis and design,

    M. Lu, S. Dutta, V . Purba, S. Dhople, and B. Johnson, “A grid- compatible virtual oscillator controller: Analysis and design,” in2019 IEEE Energy Conversion Congress and Exposition (ECCE). IEEE, 2019, pp. 2643–2649

    work page 2019

  6. [6]

    Power self-synchronization control of grid-forming voltage-source converters against a wide range of short-circuit ratio,

    P. Wang, J. Ma, R. Zhang, S. Wang, T. Liu, Z. Wu, and R. Wang, “Power self-synchronization control of grid-forming voltage-source converters against a wide range of short-circuit ratio,”IEEE Transactions on Power Electronics, vol. 38, no. 12, pp. 15 419–15 432, 2023

    work page 2023

  7. [7]

    Cross-forming control and fault current limiting for grid-forming inverters,

    X. He, M. A. Desai, L. Huang, and F. D ¨orfler, “Cross-forming control and fault current limiting for grid-forming inverters,”IEEE Transactions on Power Electronics, 2024

    work page 2024

  8. [8]

    A Data-Driven Optimal Control Architecture for Grid-Connected Power Converters

    R. Leng, L. Huang, H. Xin, P. Ju, X. Wang, E. Prieto-Araujo, and F. D ¨orfler, “DeePConverter: A data-driven optimal control architecture for grid-connected power converters,”arXiv preprint arXiv:2508.08578, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    Kundur, N

    P. Kundur, N. J. Balu, and M. G. Lauby,Power system stability and control. McGraw-hill New York, 1994, vol. 7

    work page 1994

  10. [10]

    Pai,Energy function analysis for power system stability

    A. Pai,Energy function analysis for power system stability. Springer Science & Business Media, 1989

    work page 1989

  11. [11]

    Algorithmic construc- tion of lyapunov functions for power system stability analysis,

    M. Anghel, F. Milano, and A. Papachristodoulou, “Algorithmic construc- tion of lyapunov functions for power system stability analysis,”IEEE Transactions on Circuits and Systems I: Regular Papers, vol. 60, no. 9, pp. 2533–2546, 2013

    work page 2013

  12. [12]

    Equilibrium-independent stability analysis for distribution systems with lossy transmission lines,

    W. Cui and B. Zhang, “Equilibrium-independent stability analysis for distribution systems with lossy transmission lines,”IEEE Control Sys- tems Letters, vol. 6, pp. 3349–3354, 2022

    work page 2022

  13. [13]

    Dissipativity reinforcement in feedback systems and its application to expanding power systems,

    K. Urata and M. Inoue, “Dissipativity reinforcement in feedback systems and its application to expanding power systems,”International Journal of Robust and Nonlinear Control, vol. 28, no. 5, pp. 1528–1546, 2018

    work page 2018

  14. [14]

    A passivity-based approach to voltage stabilization in DC microgrids with ZIP loads,

    P. Nahata, R. Soloperto, M. Tucci, A. Martinelli, and G. Ferrari-Trecate, “A passivity-based approach to voltage stabilization in DC microgrids with ZIP loads,”Automatica, vol. 113, p. 108770, 2020

    work page 2020

  15. [15]

    Structured neural-PI control with end-to-end stability and output tracking guarantees,

    W. Cui, Y . Jiang, B. Zhang, and Y . Shi, “Structured neural-PI control with end-to-end stability and output tracking guarantees,”Advances in Neural Information Processing Systems, vol. 36, pp. 68 434–68 457, 2023

    work page 2023

  16. [16]

    Dissipativity, stability, and connections: Progress in complexity,

    D. J. Hill and T. Liu, “Dissipativity, stability, and connections: Progress in complexity,”IEEE Control Systems Magazine, vol. 42, no. 2, pp. 88–106, 2022

    work page 2022

  17. [17]

    Interconnection of (Q, S, R)-dissipative systems in discrete time,

    A. Martinelli, A. Aboudonia, and J. Lygeros, “Interconnection of (Q, S, R)-dissipative systems in discrete time,”arXiv preprint arXiv:2311.08088, 2023

    work page arXiv 2023

  18. [18]

    Dissipativity-based data-driven decentralized control of interconnected systems,

    T. Nakano, A. Aboudonia, J. Eising, A. Martinelli, F. D ¨orfler, and J. Lygeros, “Dissipativity-based data-driven decentralized control of interconnected systems,”arXiv preprint arXiv:2509.14047, 2025

    work page arXiv 2025

  19. [19]

    Lyapunov functions family approach to transient stability assessment,

    T. L. Vu and K. Turitsyn, “Lyapunov functions family approach to transient stability assessment,”IEEE Transactions on Power Systems, vol. 31, no. 2, pp. 1269–1277, 2015

    work page 2015

  20. [20]

    Neural Lyapunov control,

    Y .-C. Chang, N. Roohi, and S. Gao, “Neural Lyapunov control,”Ad- vances in neural information processing systems, vol. 32, 2019

    work page 2019

  21. [21]

    Neural Lyapunov control for power system transient stability: A deep learning-based approach,

    T. Zhao, J. Wang, X. Lu, and Y . Du, “Neural Lyapunov control for power system transient stability: A deep learning-based approach,”IEEE Transactions on Power Systems, vol. 37, no. 2, pp. 955–966, 2021

    work page 2021

  22. [22]

    Neural networks based Lyapunov functions for transient stability analysis and assessment of power systems,

    T. Wang, X. Wang, G. Liu, Z. Wang, and Q. Xing, “Neural networks based Lyapunov functions for transient stability analysis and assessment of power systems,”IEEE Transactions on Industry Applications, vol. 59, no. 2, pp. 2626–2638, 2022

    work page 2022

  23. [23]

    Physics-informed neural networks for phase locked loop transient stability assessment,

    R. Nellikkath, I. Murzakhanov, S. Chatzivasileiadis, A. Venzke, and M. K. Bakhshizadeh, “Physics-informed neural networks for phase locked loop transient stability assessment,”Electric Power Systems Research, vol. 236, p. 110790, 2024

    work page 2024

  24. [24]

    Neural Lyapunov control of unknown nonlinear systems with stability guarantees,

    R. Zhou, T. Quartz, H. De Sterck, and J. Liu, “Neural Lyapunov control of unknown nonlinear systems with stability guarantees,”Advances in Neural Information Processing Systems, vol. 35, pp. 29 113–29 125, 2022

    work page 2022

  25. [25]

    H. K. Khalil and J. W. Grizzle,Nonlinear systems. Prentice hall Upper Saddle River, NJ, 2002, vol. 3

    work page 2002

  26. [26]

    Necessary and sufficient dissipativity-based con- ditions for feedback stabilization,

    D. d. S. Madeira, “Necessary and sufficient dissipativity-based con- ditions for feedback stabilization,”IEEE Transactions on Automatic Control, vol. 67, no. 4, pp. 2100–2107, 2021

    work page 2021

  27. [27]

    QSR-dissipativity-based stabilization of non-passive nonlinear discrete-time systems by linear static output feedback,

    T. A. Lima, D. d. S. Madeira, and M. Jungers, “QSR-dissipativity-based stabilization of non-passive nonlinear discrete-time systems by linear static output feedback,”IEEE Control Systems Letters, 2024

    work page 2024

  28. [28]

    Learning neural controllers with optimality and stability guarantees using input- output dissipativity,

    H. Wang, K. Miao, D. Madeira, and A. Papachristodoulou, “Learning neural controllers with optimality and stability guarantees using input- output dissipativity,”arXiv preprint arXiv:2506.06564, 2025

    work page arXiv 2025

  29. [29]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014. 24th Power Systems Computation Conference PSCC 2026 Limassol, Cyprus — June 8-12, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2014