Pith. sign in

REVIEW 4 major objections 4 minor 28 references

Network Interdependency-Informed Power System Dynamic Trajectory Prediction Utilizing Black-Box Modeling of Multiple Inverter-Based Resources

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper tries to establish that jointly training separate black-box predictors for every inverter-based resource with a physics-informed loss based on a linearized AC power flow yields more accurate and noise-resilient post-fault traject

desk verdict A real methods contribution with a plausible central claim, but the physics-loss term is biased at the true solution and the paper never quantifies that bias; worth serious review, conditional on code/data and additional analysis. read the letter →

arxiv 2607.05843 v3 pith:EHX5F3GY submitted 2026-07-07 eess.SY cs.SY

Sungjoo Chung , Ying Zhang , Meng Yue , Hantao Cui This is my paper · ORCID
classification eess.SYcs.SY
keywords Dynamictrajectorypredictioninverter-basedresourcesblack-boxmodelingphysics-informedneuralnetworksmachinelearningdata-drivenspatiotemporalattentionnetworkdecoupledlinearizedACpowerflow
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper addresses a practical problem: power grids with many inverter-based resources cannot be modeled from first principles because inverter controls are proprietary. The authors propose to learn a black-box predictor for each IBR from PMU measurements, then tie all predictors together during training by adding a physics-informed loss term that uses a decoupled linearized AC power flow to check that the predicted IBR voltages are consistent with the network equations. The central claim is that this joint, network-aware training produces trajectory predictions that are more accurate, more tolerant to PMU noise, and more adaptable to unseen contingencies than separately trained LSTM, physics-integrated, or graph-neural-network baselines. On the WECC 179-bus test system, the method reports voltage-magnitude RMSE 2.37e-3 and phase-angle RMSE 4.51e-3 on IBR buses, roughly 70-74% lower than the GNN baseline, and only about 13% degradation under 1% PMU noise.

What carries the argument

The central object is the modular STAN predictor: a two-layer LSTM with an attention mechanism that, for each IBR bus, reads the bus's own voltage history and the voltages of neighboring buses and outputs the next-step voltage. The carrying identity is the hybrid loss (16), L_total = L1 + λL2, with L2 built from the DL-ACPF (24), a decoupled linearization of AC power flow that expresses bus angles and magnitudes as closed-form functions of the predicted voltage-controlled-bus voltages. The DL-ACPF's analytical Jacobians (26)-(27) make the physics-informed gradient (25) computable in constant time, which is what lets all STAN predictors be updated jointly.

What would settle it

Compute the gradient of L2 at the ground-truth IBR voltages for a representative fault. If the gradient is not zero—which it cannot be unless the DL-ACPF is exact—then the physics loss pulls the predictor away from the data; testing whether this bias grows with fault severity (e.g., voltage dips below 0.9 p.u.) would settle whether the reported gains come from genuine physics or from regularization smoothing. Concretely, retrain the model with λ set to the selected value and with λ=0 on a test set of deep-sag contingencies; if the physics-informed model is worse, the DL-ACPF assumption is the

Watch

Extended reading notes

Core claim

The paper claims that the dominant source of error in multi-IBR trajectory prediction is the neglect of network interdependencies. It therefore constructs a modular spatiotemporal attention network (STAN) for each IBR bus and trains all of them end-to-end with a hybrid loss L_total = L1 + λL2, where L1 is the usual prediction error and L2 compares a DL-ACPF solution computed from the predicted IBR voltages against the ground-truth network solution. The DL-ACPF is a closed-form decoupled linearization of the AC power flow that avoids iterative solves, so gradients of L2 backpropagate efficiently. The result, the authors argue, is that predictions are pulled toward physically consistent trajec

Load-bearing premise

The load-bearing premise is that the decoupled linearized AC power-flow model is accurate enough during post-fault transients—when voltages may be far from nominal and angles large—that penalizing disagreement with its solution improves training rather than introducing a systematic bias.

Editorial extensions

If this is right

  • If the central claim is right, grid operators can predict post-fault voltage trajectories for all IBR buses simultaneously without proprietary inverter control models, using only PMU measurements and a known pre-fault network admittance matrix.
  • The method should degrade gracefully under PMU measurement errors: the reported 13% RMSE increase under 1% noise, versus larger degradations for baselines, is a direct corollary of the physics-informed regularizer.
  • Because the power-flow solution also covers non-metered buses, the same framework yields system-wide voltage trajectories, not just IBR-bus trajectories.
  • The modular design means adding or removing an IBR requires retraining only the affected predictor plus the shared loss, suggesting scalability to larger grids.
  • Unseen operating conditions, such as wider load variations or a generator trip excluded from training, are predicted with less degradation than baselines, supporting use in online stability assessment and emergency control.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves open how the method behaves when voltage departs far from nominal; since the DL-ACPF assumes near-nominal voltages and small angle differences, a deep post-fault voltage sag is the natural stress test. If the linearization error grows faster than the data loss can correct, the physics term could actively hurt accuracy.
  • The recursive inference loop feeds each STAN with DL-ACPF-estimated neighbor voltages that were themselves computed from prior predictions; the stability of this self-feeding loop is asserted, not proven. A divergence test on longer horizons (e.g., 60 s) would separate the regularizer's benefit from error accumulation.
  • A testable extension is to replace the fixed pre-fault admittance matrix with online topology identification; the paper lists this as future work, and it would also address scenarios with line switching during the fault.
  • The L2 term is not zero at the ground truth because of linearization error, so the regularizer carries a bias; systematically varying λ over a wider range, or annealing λ, might reveal a trade-off between physics consistency and data fidelity that the coarse grid search (0, 0.01, 0.05, 0.07, 0.10) does not fully map.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a modular spatiotemporal attention network (STAN) for black-box prediction of post-fault voltage trajectories at inverter-based resource (IBR) buses, trained jointly with a hybrid loss that adds a physics-informed term based on a decoupled linearized AC power flow (DL-ACPF) model. The claimed contribution is that the DL-ACPF-based regularizer enforces system-wide physical consistency and improves accuracy, especially under PMU noise and unseen scenarios. Case studies on IEEE 14-bus and WECC 179-bus systems report large RMSE reductions relative to LSTM, GNN, and PF-Integrated ML baselines, and an ablation study attributes the gains to the physics-informed and attention components.

Significance. If the reported results are correct, the method would be a practically valuable surrogate for dynamic trajectory prediction of multiple IBRs without proprietary control models, with validation-tuned regularization, a modular architecture that scales to large systems, and demonstrated robustness to measurement noise. The gradient derivation (25)-(31) is internally consistent with the stated DL-ACPF model, and the sensitivity analysis of λ in Table I is a positive feature. However, the central claim depends on the unbiasedness and accuracy of the DL-ACPF loss, and the paper currently provides only qualitative assurance on this point. The system-wide accuracy figures also conflate the STAN prediction error with the DL-ACPF linearization error. These issues are fixable with additional analysis and experiments, so the work is potentially publishable but needs substantive revision.

major comments (4)
  1. [Sec. III-B2, Eqs. (15), (16), (23)-(24)] The hybrid loss L2 is not minimized at the ground truth. When v_R equals the true IBR voltages, z_hat from (24) is the DL-ACPF solution, which differs from the true AC solution by the linearization error. Hence L2(v_R_true)=||z_hat(v_R_true)-z_true||^2>0 and ∂L2/∂v_R at the truth is nonzero; the optimizer can reduce L2 by moving v_R away from truth to compensate for DL-ACPF model error. The paper's mitigation paragraph in Sec. III-B2 is qualitative. The authors should quantify the DL-ACPF floor (L2 evaluated at ground-truth v_R) and compare it with the L2 values achieved during training, or provide an ablation that replaces the linearized term with an exact ACPF. Without this, part of the reported gains in Table I and the 'physical consistency' improvements in Table V could reflect fitting linearization bias rather than physical regularity.
  2. [Sec. III-D and Table III] The 'system-wide RMSE' reported in the last column of Table III is computed from the DL-ACPF solution z_hat, not from direct predictions of non-IBR buses. This error includes the DL-ACPF linearization error even if the STAN predictions were perfect. The paper should report an oracle baseline in which ground-truth IBR voltages are fed into the same DL-ACPF; this gives the achievable floor for system-wide accuracy. It should also clarify whether the benchmark methods use the same DL-ACPF post-processing to obtain system-wide trajectories; if they do not, the comparison is not like-for-like.
  3. [Sec. II-B, III-A, III-B2, Algorithm 1] PMUs are assumed only on IBR buses, yet each STAN uses neighboring non-IBR bus voltages y as inputs (Eq. (5)-(6)). At inference these must be obtained from the DL-ACPF estimate z_hat, creating a closed loop: STAN predictions -> DL-ACPF -> estimated y -> STAN next-step predictions. The paper does not state whether training uses ground-truth y or estimated y, nor does it analyze initialization or error propagation in this self-feeding loop. The authors should clarify the training/inference input protocol and report a comparison of ground-truth y versus estimated y inputs, since this is central to the online deployment claim.
  4. [Sec. IV-D, Table VI vs Table III] The ablation study reports a full-proposed phase-angle RMSE of 2.09e-2 under 1% PMU noise (Table VI), but Table III reports an IBR-bus phase-angle wErr RMSE of 4.68e-3 for the same system and noise level. This is a factor-of-4.5 discrepancy that is too large to be rounding. Since the ablation is the main evidence that the physics and attention components are responsible for the gains, this inconsistency must be resolved (e.g., a typo, a different test set, or different evaluation units).
minor comments (4)
  1. [Table III] The table header is very difficult to parse. Please clearly separate the IBR-bus metrics, the system-wide metrics, and the unseen-scenario metrics, and define each column group in the caption.
  2. [Table V] The text says physical consistency is evaluated by computing nodal active/reactive power and branch-current RMSEs 'through the network equations.' Please state explicitly whether these are the full nonlinear AC power-flow equations or the DL-ACPF equations. If DL-ACPF is used, the evaluation is partly circular with the training loss L2.
  3. [Sec. III-B2, 'Mitigation of Linearization Errors'] This paragraph is the only response to the known DL-ACPF approximation error, but it is qualitative. Given Major Comment 1, it should either be expanded with quantitative evidence or removed in favor of a concrete evaluation.
  4. [General] No code or data availability statement is provided. For reproducibility, please include a statement (or make the code available) for the WECC 179-bus experiments.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the physics regularizer is an external-model soft constraint and predictions are evaluated against ground truth.

full rationale

The derivation chain is self-contained and not circular. The STAN predictors produce IBR voltage predictions from PMU-measurement sequences; the DL-ACPF block in Eqs. (17)-(24) is adopted from the external reference [25], and the physics loss L2 in Eq. (15) compares the closed-form linearized power-flow solution z_hat with the historical/ground-truth network solution z_t. This is a genuine soft constraint, not a fitted target that is later renamed as a prediction. The data-driven loss L1 in Eq. (14) is standard supervised training, and the weighting coefficient lambda is selected on validation data (Table I), not tuned to the test set. The self-citations in the paper ([5], [11], [13], [17], [27]) appear as related work, baselines, or simulation tooling; none is load-bearing in the derivation. In particular, [17] is adapted as a benchmark, and [27] is the simulation environment, not a source of the paper's claimed result. The paper's own caveat in Sec. III-B2 that DL-ACPF assumes near-nominal voltages and moderate angle differences identifies a possible linearization bias in the physics term, but this is a correctness/robustness concern rather than circular reasoning: the prediction target is not defined in terms of the loss, and the reported comparisons are against external baselines and unseen scenarios. The physical-consistency metric in Table V is correlated with the L2 regularizer but is an AC-residual evaluation, not the identical training objective. Overall, the central claim retains independent empirical content.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities; all machinery is a training scheme over known components. External inputs: DL-ACPF from [25], droop-control IBR model from [23], ANDES simulator [27]. Practical assumptions: IBR-only PMU observability and a fixed pre-fault admittance matrix with post-fault effects learned from data. The only tuned scalar affecting the central claim is λ (validation-selected); network size, noise level, and window length are experimental design choices.

free parameters (4)
  • Regularization coefficient λ = 0.07 (IEEE 14-bus), 0.05 (WECC 179-bus)
    Balances data-driven loss L1 vs physics-informed loss L2 in (16); selected by validation-based grid search (Table I). Tuned on validation, but it is the scalar that controls how much the DL-ACPF regularizer shapes the central result.
  • STAN architecture hyperparameters = LSTM 128/64 units, attention 64 units, Adam lr 0.001, 100 epochs
    Hand-set architecture/training choices (Sec. IV); not fit to the test set but influence all reported accuracy numbers.
  • Input window length T_P = 120 timesteps (2.0 s at 60 Hz)
    Design choice determining how much history each predictor sees and when recursive prediction begins (Sec. IV).
  • PMU noise level = 1% maximum error
    Experimental noise setting from [28]; defines the robustness claim but is an input scenario, not a fitted constant.
assumptions (6)
  • standard math Power system dynamics are governed by nonlinear DAEs (1)-(2) with differential states for dynamic components and algebraic network equations.
    Standard modeling premise for the entire trajectory-prediction problem; invoked in Sec. II.
  • domain assumption The decoupled linearized AC power-flow model (17)-(24) from [25] approximates the true AC power flow; it assumes near-nominal voltage magnitudes and moderate voltage angle differences.
    Explicitly acknowledged in Sec. III-B2. This is the weakest supported premise because it is invoked exactly in the post-fault transient regime where those assumptions are most strained.
  • domain assumption IBR buses can be treated as voltage-controlled (PV-type) buses in the power-flow partitioning (set R), with grid-forming droop control holding the voltage.
    Sec. III-B2 and Fig. 1; the entire DL-ACPF solution (23)-(24) depends on classifying IBR buses into R with known v_R from STAN.
  • domain assumption The pre-fault nodal admittance matrix is known to the operator and fixed; post-fault effects are learned from trajectory data rather than from an updated topology.
    Sec. III-B2 states the method 'does not require an up-to-date post-fault admittance matrix.' If topology changes during the fault persistently, the fixed-matrix DL-ACPF is misspecified.
  • domain assumption PMUs are installed only on key nodes (IBR buses); neighboring-bus voltages are available in training and estimated via DL-ACPF during recursive inference.
    Sec. II-B; the recursive self-feeding of neighbor-voltage estimates is load-bearing for the multi-step prediction scheme (6).
  • domain assumption Droop control is representative of grid-forming inverter control; extension to other control types is deferred to future work.
    Sec. II-A explicitly frames the paper as a proof-of-concept on droop control; the trained black boxes may not transfer to other control modes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Network Interdependency-Informed Power System Dynamic Trajectory Prediction Utilizing Black-Box Modeling of Multiple Inverter-Based Resources." pith.science (2026). https://pith.science/paper/EHX5F3GY

@misc{pith2026260705843,
  author       = {Pith},
  title        = {Pith review of: Network Interdependency-Informed Power System Dynamic Trajectory Prediction Utilizing Black-Box Modeling of Multiple Inverter-Based Resources},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EHX5F3GY}},
  note         = {Machine review of arXiv:2607.05843}
}
read the original abstract

Black-box modeling of inverter-based resources (IBRs) has attracted growing interest for real-time grid operation and control in the presence of proprietary electronic control architectures. Existing machine learning (ML)-based online dynamic trajectory prediction approaches using IBR black-box models either significantly accumulate prediction errors when multiple surrogates are simultaneously used or ignore measurement errors, limiting their deployment in practical grids. To address these limitations, this paper proposes a novel network interdependency-informed ML algorithm for online dynamic trajectory prediction in IBR-integrated power systems. A modular spatiotemporal attention network (STAN)-based predictor for the black-box modeling of each IBR unit is first proposed. Utilizing past measurements, the proposed STAN can effectively capture and predict the spatiotemporal dynamics of IBRs by employing an attention mechanism to attend to the most pertinent features for trajectory prediction. Furthermore, a novel hybrid physics-informed loss function that integrates a decoupled linearized AC power flow formulation is proposed. The proposed loss function effectively ensures physical consistency of predictions within network operation while avoiding the computational complexity of iterative power flow solving, thereby enabling efficient gradient backpropagation and overall improved prediction accuracy. Case studies on the IEEE 14- and WECC 179-bus systems demonstrate that the proposed method achieves significant accuracy enhancement and robustness against measurement errors, outperforming recent ML-based trajectory prediction methods.

Figures

Figures reproduced from arXiv: 2607.05843 by the authors.

Figure 2
Figure 2. Diagram of black-box modeling of a single inverter-based [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Proposed STAN-based predictor architecture for IBR bus [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. 16 randomly selected voltage magnitude trajectories of an IBR [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figures from the paper (6 more)
Figure 7
Figure 7. Figure 7: Trajectory prediction comparison on IBR 3 in the WECC [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 6
Figure 6. Figure 6: Per-trajectory RMSE distribution comparison of all the IBR [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: Trajectory prediction comparison on multiple IBR buses in [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 7
Figure 7. Figure 7: Trajectory prediction comparison between the proposed [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Trajectory prediction comparison between the proposed [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: illustrates the detailed results of voltage magnitude and phase angle trajectories on an IBR bus. The purely data￾driven STAN and physics-informed LSTM variants exhibit larger deviations from the actual response than the full pro￾posed method. These trajectory-level re…

Review history (3 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references

  1. [1]

    Grid- forming converters: Control approaches, grid-synchronization, and future trends—a review,

    R. Rosso, X. Wang, M. Liserre, X. Lu, and S. Engelken, “Grid- forming converters: Control approaches, grid-synchronization, and future trends—a review,”IEEE Open Journal of Industry Applications, vol. 2, pp. 93–109, 2021

    2021

  2. [2]

    A more resilient grid: The u.s. department of energy joins with stakeholders in an r&d plan,

    D. T. Ton and W.-T. P. Wang, “A more resilient grid: The u.s. department of energy joins with stakeholders in an r&d plan,”IEEE Power and Energy Magazine, vol. 13, no. 3, pp. 26–34, 2015

    2015

  3. [3]

    Data- driven dynamic modeling in power systems: A fresh look on inverter- based resource modeling,

    L. Fan, Z. Miao, S. Shah, P. Koralewicz, V . Gevorgian, and J. Fu, “Data- driven dynamic modeling in power systems: A fresh look on inverter- based resource modeling,”IEEE Power and Energy Magazine, vol. 20, no. 3, pp. 64–76, 2022

    2022

  4. [4]

    An online power system stability monitoring system using convolutional neural networks,

    A. Gupta, G. Gurrala, and P. S. Sastry, “An online power system stability monitoring system using convolutional neural networks,”IEEE Transactions on Power Systems, vol. 34, no. 2, pp. 864–872, 2019

    2019

  5. [5]

    Multi-agent graph-attention deep reinforcement learning for post-contingency grid emergency volt- age control,

    Y . Zhang, M. Yue, J. Wang, and S. Yoo, “Multi-agent graph-attention deep reinforcement learning for post-contingency grid emergency volt- age control,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 3, pp. 3340–3350, 2024

    2024

  6. [6]

    A model predictive approach for enhancing transient stability of grid-forming converters,

    A. Arjomandi-Nezhad, Y . Guo, B. C. Pal, and D. Varagnolo, “A model predictive approach for enhancing transient stability of grid-forming converters,”IEEE Transactions on Power Systems, vol. 39, no. 5, pp. 6675–6688, 2024

    2024

  7. [7]

    Efficient transient stability analysis of electrical power system based on a spatially paral- leled hybrid approach,

    S. Xia, S. Bu, J. Hu, B. Hong, Z. Guo, and D. Zhang, “Efficient transient stability analysis of electrical power system based on a spatially paral- leled hybrid approach,”IEEE Transactions on Industrial Informatics, vol. 15, no. 3, pp. 1460–1473, 2019

    2019

  8. [8]

    A multi-decomposition approach for accelerated time-domain simulation of transient stability problems,

    P. Zadkhast, J. Jatskevich, and E. Vaahedi, “A multi-decomposition approach for accelerated time-domain simulation of transient stability problems,”IEEE Transactions on Power Systems, vol. 30, no. 5, pp. 2301–2311, 2015

    2015

Show all 28 references
  1. [9]

    Ieee standard for interconnection and interoperability of inverter-based resources (ibrs) interconnecting with associated transmission electric power systems,

    “Ieee standard for interconnection and interoperability of inverter-based resources (ibrs) interconnecting with associated transmission electric power systems,”IEEE Std 2800-2022, pp. 1–180, 2022

    2022

  2. [10]

    Physics-informed neural networks: A plug and play integration into power system dynamic simulations,

    I. Ventura Nadal, J. Stiasny, and S. Chatzivasileiadis, “Physics-informed neural networks: A plug and play integration into power system dynamic simulations,”Electric Power Systems Research, vol. 248, p. 111885, Nov. 2025

    2025

  3. [11]

    Machine-learning-based online transient analysis via iterative computation of generator dynamics,

    J. Li, M. Yue, Y . Zhao, and G. Lin, “Machine-learning-based online transient analysis via iterative computation of generator dynamics,” in 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), 2020, pp. 1–6

    2020

  4. [12]

    The use of machine learning for prediction of post-fault rotor angle trajectories,

    X. Ye, A. Radovanovi ´c, and J. V . Milanovi ´c, “The use of machine learning for prediction of post-fault rotor angle trajectories,”IEEE Transactions on Power Systems, vol. 39, no. 5, pp. 6496–6507, 2024

    2024

  5. [13]

    Structure-informed graph learning of networked dependencies for online prediction of power system transient dynamics,

    T. Zhao, M. Yue, and J. Wang, “Structure-informed graph learning of networked dependencies for online prediction of power system transient dynamics,”IEEE Transactions on Power Systems, vol. 37, no. 6, pp. 4885–4895, 2022

    2022

  6. [14]

    A frequency domain approach to predict power system transients,

    W. Cui, W. Yang, and B. Zhang, “A frequency domain approach to predict power system transients,”IEEE Transactions on Power Systems, vol. 39, no. 1, pp. 465–477, 2024

    2024

  7. [15]

    Bayesian post-fault power system dynamic trajec- tory prediction,

    B. Tan and J. Zhao, “Bayesian post-fault power system dynamic trajec- tory prediction,”IEEE Transactions on Power Systems, pp. 1–13, 2025

    2025

  8. [16]

    Data-driven modeling for grid edge ibrs: A digital twin perspective of user-defined models,

    K. Mahapatra, B. Mitra, and S. Kundu, “Data-driven modeling for grid edge ibrs: A digital twin perspective of user-defined models,” inIECON 2024 - 50th Annual Conference of the IEEE Industrial Electronics Society, 2024, pp. 1–7

    2024

  9. [17]

    Integrating learning and physics based computation for fast online transient analysis,

    J. Li, Y . Zhao, and M. Yue, “Integrating learning and physics based computation for fast online transient analysis,” in2023 IEEE Power & Energy Society Innovative Smart Grid Technologies Conference (ISGT), 2023, pp. 1–5

    2023

  10. [18]

    Online dynamic security assessment: Using hybrid physics-guided deep learning models,

    G. Lu and S. Bu, “Online dynamic security assessment: Using hybrid physics-guided deep learning models,”IEEE Transactions on Industrial Informatics, vol. 20, no. 11, pp. 13 190–13 200, 2024

    2024

  11. [19]

    Advanced probabilistic transient stability assess- ment for operational planning: A physics-informed graphical learning approach,

    G. . Lu and S. Bu, “Advanced probabilistic transient stability assess- ment for operational planning: A physics-informed graphical learning approach,”IEEE Transactions on Power Systems, vol. 40, no. 1, pp. 740–752, 2025

    2025

  12. [20]

    Gray-box modeling for distribution systems with inverter-based resources: Integrating physics- based and data-driven approaches,

    J. Zhang, Y . Men, L. Ding, X. Lu, and W. Du, “Gray-box modeling for distribution systems with inverter-based resources: Integrating physics- based and data-driven approaches,”IEEE Transactions on Industry Applications, vol. 60, no. 4, pp. 5490–5498, 2024

    2024

  13. [21]

    Time-domain measurement-baseddq-frame admit- tance model identification for inverter-based resources,

    L. Fan and Z. Miao, “Time-domain measurement-baseddq-frame admit- tance model identification for inverter-based resources,”IEEE Transac- tions on Power Systems, vol. 36, no. 3, pp. 2211–2221, 2021

    2021

  14. [22]

    Learning to model the dynamics of black-box inverter-based resources with multiple unknown control modes from noisy measurement data,

    H. Huang and Y . Lin, “Learning to model the dynamics of black-box inverter-based resources with multiple unknown control modes from noisy measurement data,”IEEE Transactions on Smart Grid, vol. 17, no. 3, pp. 2530–2543, 2026. xiii

    2026

  15. [23]

    Modeling of grid-forming and grid-following invert- ers for dynamic simulation of large-scale distribution systems,

    W. Du, F. K. Tuffner, K. P. Schneider, R. H. Lasseter, J. Xie, Z. Chen, and B. Bhattarai, “Modeling of grid-forming and grid-following invert- ers for dynamic simulation of large-scale distribution systems,”IEEE Transactions on Power Delivery, vol. 36, no. 4, pp. 2035–2045, 2021

  16. [24]

    Online identification of power system dynamic signature using pmu measurements and data mining,

    T. Guo and J. V . Milanovi ´c, “Online identification of power system dynamic signature using pmu measurements and data mining,”IEEE Transactions on Power Systems, vol. 31, no. 3, pp. 1760–1768, 2016

    2016

  17. [25]

    A state-independent linear power flow model with accurate estimation of voltage magnitude,

    J. Yang, N. Zhang, C. Kang, and Q. Xia, “A state-independent linear power flow model with accurate estimation of voltage magnitude,”IEEE Transactions on Power Systems, vol. 32, no. 5, pp. 3607–3617, 2017

    2017

  18. [26]

    A test cases library for methods locating the sources of sustained oscillations,

    S. Maslennikov, B. Wang, Q. Zhang, a. Ma, a. Luo, a. Sun, and E. Litvinov, “A test cases library for methods locating the sources of sustained oscillations,” in2016 IEEE Power and Energy Society General Meeting (PESGM), 2016, pp. 1–5

    2016

  19. [27]

    Hybrid symbolic-numeric framework for power system modeling and analysis,

    H. Cui, F. Li, and K. Tomsovic, “Hybrid symbolic-numeric framework for power system modeling and analysis,”IEEE Transactions on Power Systems, vol. 36, no. 2, pp. 1373–1384, 2021

    2021

  20. [28]

    Attack identification and correction for pmu gps spoofing in unbalanced distribution systems,

    Y . Zhang, J. Wang, and J. Liu, “Attack identification and correction for pmu gps spoofing in unbalanced distribution systems,”IEEE Transac- tions on Smart Grid, vol. 11, no. 1, pp. 762–773, 2020

    2020

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.