Pith. sign in

REVIEW 5 major objections 7 minor 63 references

Towards Robust Stability Prediction in Smart Grids: GAN-based Approach under Data Constraints and Adversarial Challenges

T0 review · 5 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a GAN trained only on stable grid data can detect instability and adversarial attacks, replacing the need for real unstable samples during training.

desk verdict Useful one-class GAN framework for smart grid stability with honest reporting, but the key assumption that OOD samples represent real instability is untested and the Algorithm 1 sign needs checking. read the letter →

arxiv 2501.16490 v2 pith:V4HT43SI submitted 2025-01-27 cs.CR cs.AIcs.LG

Emad Efatinasab , Alessandro Brighente , Denis Donadel , Mauro Conti , Mirco Rampazzo This is my paper · ORCID
classification cs.CRcs.AIcs.LG
keywords SmartgridsStabilitypredictionGenerativeadversarialnetworksOut-of-distributiondetectionattacksOne-classclassificationDatascarcity
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 claims that smart-grid instability can be detected without ever training on real unstable samples. It builds a GAN whose generator is deliberately kept from mimicking the stable class, instead producing out-of-distribution points that stand in for failures, anomalies, and attacks. A discriminator trained on stable data plus these synthetic outliers plus FGSM adversarial examples reaches 98.1 percent accuracy on the augmented UCI stability dataset and flags adversarial perturbations as unstable at up to 98.9 percent accuracy. If this holds, one-class training could replace expensive collection of instability data while also absorbing adversarial attacks into the same model.

What carries the argument

The load-bearing object is the repulsion-loss GAN: the generator minimizes -E[log D(G(z))] plus E[ReLU(m - ||x - s||)] against stable samples s, forcing generated points into X_OOD = {x : ||x - s|| >= m for all s in S}. This turns the generator into an oracle for plausible non-stable regions without real failure data, and the FGSM adversarial layer injects perturbed stable points as an extra 'unstable' class, making the discriminator both a stability predictor and an attack detector.

What would settle it

Take the real unstable samples held out from training and measure the distance from each generated OOD sample to its nearest real unstable neighbor; if generated samples are no closer to real unstable points than to random points, or if a discriminator trained only on stable data and OOD samples scores near chance on real unstable samples from a second grid topology, then the OOD-to-instability equivalence is disproved.

Watch

Extended reading notes

Core claim

GAN-Stability is a one-class stability classifier: the discriminator is the final predictor, trained only on real stable samples, generator-produced out-of-distribution samples, and FGSM-perturbed stable samples labeled unstable. The generator is held away from the stable manifold by a repulsion loss with margin m=4 (four standard deviations after z-score normalization), so the discriminator learns a boundary around the stable region instead of a boundary between two real classes. The authors report that this boundary separates stable from unstable test samples with mean accuracy 0.981, detects state-of-the-art white-box attacks as unstable with mean accuracy 0.989 (0.988 grey-box, 0.991 for GAN-GRID), and runs on a Raspberry Pi in under 7 ms per batch.

Load-bearing premise

The whole approach assumes that the generator's synthetic out-of-distribution points, kept at least four standard deviations from stable data, actually resemble real unstable grid behavior rather than arbitrary noise; the paper gives no direct measurement of that resemblance.

Editorial extensions

If this is right

  • Stability models could be trained from normal-operation readings alone, removing the need to induce or collect rare unstable states.
  • The same discriminator replaces a separate anomaly-detection system, since adversarial perturbations are classified as instability and raise an alert.
  • Deployment on low-cost hardware with sub-10 ms response makes per-node monitoring feasible in decentralized smart grids.
  • Accuracy around 0.98 is reached while using only about one-third of the dataset for training, suggesting strong sample efficiency.

Reading between the lines

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

  • The repulsion margin m=4 is anchored to the empirical rule for the stable class; on a non-Gaussian or multimodal stable distribution, the same margin may either enclose large stable regions or push generated samples into nonsensical parts of the input space, so the margin may need to be data-dependent.
  • Because attacks are identified as 'unstable,' the model cannot distinguish an adversarial perturbation of a stable reading from a genuinely dangerous grid state; a follow-up could separate 'attack' from 'fault' using a second decision layer.
  • The transferability argument is doing a lot of work: FGSM training is claimed to generalize to BIM, RFGSM, PGD, and GAN-GRID detections, so a testable extension is to train with only one attack type per model and measure cross-attack detection as a function of epsilon.
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

5 major / 7 minor

Summary. The paper proposes GAN-Stability, a GAN-based framework for smart grid stability prediction trained exclusively on stable instances. The generator is trained with a repulsion loss to produce out-of-distribution (OOD) samples relative to the stable class, and the discriminator learns to separate stable samples from these synthetic OOD samples. An adversarial training layer augments the discriminator with FGSM-generated examples labeled as unstable, enabling attack detection within the same model. The method is evaluated on the augmented UCI Electrical Grid Stability Simulated Dataset, reporting 0.981 accuracy for stability prediction, 0.989 white-box and 0.988 grey-box detection of adversarial attacks, and sub-7 ms inference on a Raspberry Pi 4. The authors benchmark against existing supervised models and release code and data.

Significance. If the central assumption holds, the paper addresses a practical gap: stability prediction without recourse to rare and expensive unstable-instance labels, combined with integrated adversarial-attack detection. The reported hardware deployment and open-source release are strengths that increase reproducibility and practical relevance. However, the significance depends critically on whether the generator's OOD samples are representative of real unstable grid behavior; the paper provides no direct evidence for this. The lack of one-class baselines and the absence of error bars further weaken the empirical claims. With additional validation, the approach could be a useful contribution to one-class stability monitoring in smart grids, but in its current form the empirical case is incomplete.

major comments (5)
  1. [§5.2, Eq. (7)–(8)] The load-bearing assumption that the generator's OOD samples represent unstable grid states is not established. The repulsion loss in Eq. (7) only enforces that generated samples are at least m away from stable training samples, and the set XOOD in Eq. (8) is defined purely by this distance condition. The statement in §5.2 that 'adversarial pressure from the discriminator ensures these samples remain realistic' is not supported by the algorithm: Step 2 of Algorithm 1 trains the discriminator to label generated samples as unstable, not to judge physical realism or manifold membership. The reported accuracy on the simulated UCI dataset is compatible with a one-class boundary around the stable class that separates arbitrary off-manifold noise from stable data, without transferring to actual instability. The authors should provide direct evidence that generated samples resemble held-out unstable samples, for example through density comparisons, visualizations in a low-dimensional embedding, or evaluation of a supervised classifier trained on generated samples against true unstable labels.
  2. [Algorithm 1, Step 4 vs. §5.2, Eq. (7)] There is a sign inconsistency in the generator objective. Eq. (7) defines LG = -E_z[log D(G(z))] + repulsion, whereas Algorithm 1 Step 4 writes LG = E[log(D(Xfake))] + Lrepulsion and then backpropagates to update G. These two objectives push the generator in opposite directions with respect to the discriminator's stable/unstable output. Since the released code is the authoritative implementation, the authors should correct one of the two formulations and explicitly state which objective was used in the reported experiments.
  3. [§6.3, Tables 2 and 3] All accuracy and F1 values are presented as point estimates with no error bars, confidence intervals, or multiple-seed reporting, despite the text repeatedly calling them 'mean accuracies.' The improvements from 0.918 to 0.981 for stability prediction and the attack-detection numbers in Table 2 therefore lack statistical support. The authors should report mean and standard deviation over at least 5–10 independent training runs and, where relevant, a significance test over paired evaluations.
  4. [§6.4 and §6.2] The paper does not compare against standard one-class classification baselines such as one-class SVM, isolation forest, autoencoder-based anomaly detection, or Deep SVDD. Since GAN-Stability is fundamentally a one-class classifier trained only on stable data, these baselines are necessary to support the claim that the GAN's OOD generation improves boundary learning. Without such comparisons, it is unclear whether the generator adds value over a simpler one-class boundary on the same stable training data.
  5. [§5.3 and §6.3] The evaluation of the adversarial training layer is partially circular. The adversarial training in Algorithm 1 uses FGSM on the stable training data with epsilon = 0.05, and the white-box evaluation in Table 2 includes FGSM with the same epsilon. The perfect 1.000 FGSM detection accuracy is therefore expected and does not by itself demonstrate generalization. The high transfer accuracies for BIM, RFGSM, and PGD are more informative, but the authors should also evaluate on attack types and perturbation sizes not seen during training to support the robustness claim.
minor comments (7)
  1. [Abstract and §6.1] The dataset is described as 'real-world' in the abstract and Section 6.1, but the Electrical Grid Stability Simulated Dataset is a simulation-based dataset. Please revise to 'publicly available simulated dataset' or similar.
  2. [§5.2 and §6.0] Equation numbers are reused: Eq. (8) defines XOOD in Section 5.2, and Eq. (8) defines accuracy in Section 6. Renumber to avoid confusion.
  3. [§5.2] The norm in Eq. (5) is not specified. State whether ||x - s|| is the Euclidean norm computed on z-scored features, and clarify the dimension of the margin m.
  4. [§5.2] The justification of m = 4 via the Empirical Rule is not rigorous: the Empirical Rule applies to univariate normal distributions, while m = 4 is a distance threshold in a 12-dimensional feature space. Clarify the reasoning or provide an empirical sensitivity study.
  5. [Algorithm 1, Step 3] The FGSM update writes Xadv = Xreal + epsilon * sign(gradient of LD_real), but the text does not state whether the gradient is taken with respect to the input batch Xreal or with respect to the model parameters. Clarify the notation.
  6. [Table 3] F1 scores for the stable and unstable classes are listed as '-'; these values should be reported or explicitly omitted for clarity.
  7. [§6.5] The Kaggle environment is described as having an Intel Xeon CPU, but the training time comparison would benefit from specifying whether a GPU was used and, if so, its model.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity in the white-box FGSM attack-detection metric; the central stability-prediction result is independent and self-contained.

  1. fitted input called prediction [Section 5.3 (Adversarial Training), Algorithm 1 Step 3; Section 6.3 and Table 2]
    "Step 3 (adversarial samples) ... Generate adversarial samples using FGSM: Xadv = Xreal + ϵFGSM · sign(∇XLD real) ... Compute discriminator loss for adversarial data: LD adv = E[log(1 − D(Xadv))] ... The discriminator is then trained on these data with unstable labels, which can enhance its ability to detect FGSM samples as instances of instability within the grid. ... We subject the discriminator model and testing dataset to various attacks, including FGSM ... With a mean accuracy of 0.989, our system adeptly identifies these attacks as belonging to the unstable class."

    The white-box FGSM test is not a test on an unseen attack family: the discriminator was trained, in Algorithm 1 Step 3, on FGSM-perturbed stable samples labeled unstable, using the same FGSM transform computed against the same discriminator. The evaluation in Section 6.3 then crafts FGSM attacks against that discriminator and reports their classification accuracy as attack-detection performance. The reported FGSM score (1.000 in Table 2) is therefore a fidelity check of the training objective — the model predicting on the same input distribution it was trained to reject — rather than evidence of generalizing to a held-out attack type. This inflates the headline white-box mean of 0.989 by construction.

full rationale

The central stability-prediction claim is not circular: Section 6.1 reserves 10% of stable samples and all unstable samples for testing, and the model is trained only on 90% of the stable class plus synthetic OOD and FGSM samples. The reported 0.981 accuracy on real held-out unstable labels is therefore an independent evaluation of the learned boundary. The Section 5.2 assumption that repulsion-generated samples 'converge to high-density regions within this complement' and represent plausible unstable states is unproven and is a legitimate external-validity concern, but it is not circular: the test labels are not used to set m or to train the generator. The m=4 margin is grounded in stable-class statistics, not in test outcomes. The one genuine circularity is narrower and secondary: white-box FGSM detection is trained on exactly the FGSM transform it later reports as a detection result. Because the main stability benchmark, the grey-box transfer attacks, and the GAN-GRID scenario all remain independent, the paper's central claim is not forced; the partial circularity affects only the white-box FGSM component of the adversarial-detection headline. A separate code/math inconsistency (Algorithm 1 Step 4 writes LG = E[log(D(Xfake))] + Lrepulsion, omitting the negative sign of Eq. (7)) is a correctness concern, not a circularity, and was not counted in the score.

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

The central claim rests on several unproven or hand-tuned elements: the margin m=4, the convergence of the generator to useful OOD regions, transferability of FGSM training to other attacks, and the validity of the simulated dataset as a proxy for real grids. No new physical entities are introduced.

free parameters (4)
  • Repulsion margin m = 4 (z-score normalized units)
    Section 5.2: chosen from the empirical rule and empirically validated on the stable class; directly sets the OOD exclusion zone B_m(S) in Eq. (8).
  • Adversarial perturbation bound epsilon = 0.05
    Section 6.3: used for FGSM adversarial training and for evaluating FGSM, BIM, RFGSM, and PGD; selected as a balance without sensitivity analysis.
  • GAN training schedule = 250 epochs, learning rate 0.0002, batch size 4
    Section 5.2: hand-set hyperparameters; the generator is intentionally not converged, making epoch count an influential stopping rule.
  • Generator and discriminator widths = Generator (100,128,64,12), Discriminator (160,200,256,512,1)
    Table 1: architectural asymmetry is claimed to help the discriminator dominate; no ablation supports these specific dimensions.
assumptions (6)
  • domain assumption Stable training data is comprehensive and representative of all stable grid states.
    Section 3.1 uses only stable instances for training; Section 7 acknowledges effectiveness depends on a diverse and comprehensive stable-class dataset.
  • domain assumption The augmented UCI Electrical Grid Stability Simulated Dataset is a valid proxy for real-world smart grid stability.
    Section 6.1 uses the simulated 4-node DSGC dataset; the abstract calls it real-world; Section 7 notes lack of large-scale real grid data.
  • domain assumption Stable class features are approximately normal, justifying the empirical-rule choice of m=4.
    Section 5.2 states most stable samples lie within three standard deviations and this follows the Empirical Rule; no normality test is reported.
  • ad hoc to paper The generator distribution pg converges to high-density regions inside XOOD under repulsion loss and adversarial pressure.
    Section 5.2, Eq. (8): asserted without proof or direct empirical validation that generated samples resemble real unstable behavior.
  • domain assumption FGSM-based adversarial training transfers to BIM, RFGSM, PGD, and GAN-GRID attacks.
    Section 5.3 relies on transferability properties from cited work [50] rather than a formal guarantee.
  • standard math Standard GAN optimal discriminator and generator results from Goodfellow et al. apply to this training setup.
    Section 5.2 invokes Eqs. (2)-(4); accepted background but assumes standard GAN assumptions hold despite the modified repulsion loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Robust Stability Prediction in Smart Grids: GAN-based Approach under Data Constraints and Adversarial Challenges." pith.science (2026). https://pith.science/paper/V4HT43SI

@misc{pith2026250116490,
  author       = {Pith},
  title        = {Pith review of: Towards Robust Stability Prediction in Smart Grids: GAN-based Approach under Data Constraints and Adversarial Challenges},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V4HT43SI}},
  note         = {Machine review of arXiv:2501.16490}
}
read the original abstract

Smart grids are crucial for meeting rising energy demands driven by global population growth and urbanization. By integrating renewable energy sources, they enhance efficiency, reliability, and sustainability. However, ensuring their availability and security requires advanced operational control and safety measures. Although artificial intelligence and machine learning can help assess grid stability, challenges such as data scarcity and cybersecurity threats, particularly adversarial attacks, remain. Data scarcity is a major issue, as obtaining real-world instances of grid instability requires significant expertise, resources, and time. Yet, these instances are critical for testing new research advancements and security mitigations. This paper introduces a novel framework for detecting instability in smart grids using only stable data. It employs a Generative Adversarial Network (GAN) where the generator is designed not to produce near-realistic data but instead to generate Out-Of-Distribution (OOD) samples with respect to the stable class. These OOD samples represent unstable behavior, anomalies, or disturbances that deviate from the stable data distribution. By training exclusively on stable data and exposing the discriminator to OOD samples, our framework learns a robust decision boundary to distinguish stable conditions from any unstable behavior, without requiring unstable data during training. Furthermore, we incorporate an adversarial training layer to enhance resilience against attacks. Evaluated on a real-world dataset, our solution achieves up to 98.1\% accuracy in predicting grid stability and 98.9\% in detecting adversarial attacks. Implemented on a single-board computer, it enables real-time decision-making with an average response time of under 7ms.

Figures

Figures reproduced from arXiv: 2501.16490 by the authors.

Figure 1
Figure 1. GAN-Stability general architecture. 5.1. Architecture Introduced by Goodfellow et al. [48], GANs consist of two neural networks— a generator and a discriminator—engaged in an adversarial game. The gen￾erator’s objective is to produce synthetic data samples that closely resemble 12 [PITH_FULL_IMAGE:figures/full_fig_p014_1.png] view at source ↗
Figure 2
Figure 2. Generator and Discriminator losses for the final model. [PITH_FULL_IMAGE:figures/full_fig_p019_2.png] view at source ↗
Figure 3
Figure 3. Repulsion loss for the final model. with prolonged training and repulsion loss, ensures the generator does not converge too closely to the stable data distribution. Instead, it generates OOD samples with respect to the stable class, which may be representative of any form of unstable behavior. GAN-Stability is trained over 250 epochs with a learning rate of 0.0002 and a batch size of 4, ensuring thorough learning an… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: ROC curve for classification performance of GAN-Stability augmented with the [PITH_FULL_IMAGE:figures/full_fig_p028_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 51 canonical work pages

  1. [1]

    Breviglieri, T

    P. Breviglieri, T. Erdem, S. Eken, Predicting smart grid stability with optimized deep models, SN Computer Science 2 (2021) 1–12

    work page 2021

  2. [2]

    H. A. Muqeet, R. Liaqat, M. Jamil, A. A. Khan, A state-of-the-art review of smart energy systems and their management in a smart grid environment, Energies 16 (1) (2023) 472

    work page 2023

  3. [3]

    Efatinasab, A

    E. Efatinasab, A. Brighente, M. Rampazzo, N. Azadi, M. Conti, Gan- grid: A novel generative attack on smart grid stability prediction, in: J. Garcia-Alfaro, R. Kozik, M. Chora´ s, S. Katsikas (Eds.), Computer Security – ESORICS 2024, Springer Nature Switzerland, Cham, 2024, pp. 374–393

    work page 2024

  4. [4]

    Jiao, Application and prospect of artificial intelligence in smart grid, IOP Conference Series: Earth and Environmental Science 510 (2) (2020) 022012

    J. Jiao, Application and prospect of artificial intelligence in smart grid, IOP Conference Series: Earth and Environmental Science 510 (2) (2020) 022012. doi:10.1088/1755-1315/510/2/022012. 30

    work page doi:10.1088/1755-1315/510/2/022012 2020

  5. [5]

    Albadi, E

    M. Albadi, E. El-Saadany, A summary of demand response in electricity markets, Electric Power Systems Research 78 (11) (2008) 1989–1996. doi:https://doi.org/10.1016/j.epsr.2008.04.002

    doi:10.1016/j.epsr.2008.04.002 2008

  6. [6]

    Palensky, D

    P. Palensky, D. Dietrich, Demand side management: Demand response, intelligent energy systems, and smart loads, IEEE Transactions on In- dustrial Informatics 7 (3) (2011) 381–388. doi:10.1109/TII.2011. 2158841

    work page doi:10.1109/tii.2011 2011

  7. [7]

    M. B. Gorza lczany, J. Piekoszewski, F. Rudzi´ nski, A modern data- mining approach based on genetically optimized fuzzy systems for inter- pretable and accurate smart-grid stability prediction, Energies 13 (10) (2020). doi:10.3390/en13102559

    work page doi:10.3390/en13102559 2020

  8. [8]

    Sch¨ afer, M

    B. Sch¨ afer, M. Matthiae, M. Timme, D. Witthaut, Decentral smart grid control, New journal of physics 17 (1) (2015) 015002

    work page 2015

Show all 63 references
  1. [9]

    J. A. Short, D. G. Infield, L. L. Freris, Stabilization of grid frequency through dynamic demand control, IEEE Transactions on Power Systems 22 (3) (2007) 1284–1293. doi:10.1109/TPWRS.2007.901489

    2007

  2. [10]

    Arzamasov, K

    V. Arzamasov, K. B¨ ohm, P. Jochem, Towards concise models of grid stability, in: 2018 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGrid- Comm), 2018, pp. 1–6. doi:10.1109/SmartGridComm.2018.8587498

    2018

  3. [11]

    Electrical, What happens if your electricity supply is not stable?, accessed: 2025-01-11 (n.d.)

    V. Electrical, What happens if your electricity supply is not stable?, accessed: 2025-01-11 (n.d.). URL https://www.voltfixelectrical.com.au/blog/ what-happens-if-your-electricity-supply-is-not-stable/

    2025

  4. [12]

    URL https://www.politico.com/news/2024/12/31/ puerto-rico-near-total-blackout-00196163

    Politico, Puerto rico hit by near-total blackout, accessed: 2025-01-11 (2024). URL https://www.politico.com/news/2024/12/31/ puerto-rico-near-total-blackout-00196163

    2024

  5. [13]

    Zhang, H

    Y. Zhang, H. Zhang, J. Zhang, L. Li, Z. Zheng, Power grid stabil- ity prediction model based on bilstm with attention, in: 2021 Inter- national Symposium on Electrical, Electronics and Information Engi- neering, ISEEIE 2021, Association for Computing Machinery, 2021, p. 344–34...

    2021

  6. [14]

    Alazab, S

    M. Alazab, S. Khan, S. S. R. Krishnan, Q.-V. Pham, M. P. K. Reddy, T. R. Gadekallu, A multidirectional lstm model for predict- ing the stability of a smart grid, IEEE Access 8 (2020) 85454–85463. doi:10.1109/ACCESS.2020.2991067

    2020

  7. [15]

    Massaoudi, H

    M. Massaoudi, H. Abu-Rub, S. S. Refaat, I. Chihi, F. S. Oueslati, Accu- rate smart-grid stability forecasting based on deep learning: Point and interval estimation method, in: 2021 IEEE Kansas Power and Energy Conference (KPEC), 2021, pp. 1–6. doi:10.1109/KPEC51835.2021. 9446196

    2021

  8. [16]

    W. Guo, X. Zha, K. Qian, T. Chen, Can active learning benefit the smart grid? a perspective on overcoming the data scarcity, in: 2019 IEEE 2nd International Conference on Electronics and Communication Engineering (ICECE), 2019, pp. 346–350. doi:10.1109/ICECE48499. 2019.9058539

    2019

  9. [17]

    Efatinasab, F

    E. Efatinasab, F. Marchiori, A. Brighente, M. Rampazzo, M. Conti, Faultguard: A generative approach to resilient fault prediction in smart electrical grids, in: F. Maggi, M. Egele, M. Payer, M. Carminati (Eds.), Detection of Intrusions and Malware, and Vulnerability Assessment...

    2024

  10. [18]

    Efatinasab, A

    E. Efatinasab, A. Sinigaglia, N. Azadi, G. Antonio Susto, M. Rampazzo, Adversarially robust fault zone prediction in smart grids with bayesian neural networks, IEEE Access 12 (2024) 121169–121184. doi:10.1109/ ACCESS.2024.3452476

    2024

  11. [19]

    Q. Song, R. Tan, C. Ren, Y. Xu, Y. Lou, J. Wang, H. B. Gooi, On cred- ibility of adversarial examples against learning-based grid voltage sta- bility assessment, IEEE Transactions on Dependable and Secure Com- puting 21 (2) (2024) 585–599. doi:10.1109/TDSC.2022.3213012

    2024

  12. [20]

    Q. Song, R. Tan, C. Ren, Y. Xu, Understanding credibility of ad- versarial examples against smart grid: A case study for voltage sta- bility assessment, in: Proceedings of the Twelfth ACM International Conference on Future Energy Systems, e-Energy ’21, Association for Computin...

    2021

  13. [21]

    X. Ke, D. Wu, J. Rice, M. Kintner-Meyer, N. Lu, Quantifying impacts of heat waves on power grid operation, Applied Energy 183 (2016) 504–512. doi:https://doi.org/10.1016/j.apenergy.2016.08.188. URL https://www.sciencedirect.com/science/article/pii/ S0306261916312971

    2016 doi

  14. [22]

    Ourahou, W

    M. Ourahou, W. Ayrir, B. E. Hassouni, A. Haddi, Review on smart grid control and reliability in presence of renewable energies: Challenges and prospects, Mathematics and computers in simulation 167 (2020) 19–31

    2020

  15. [23]

    S. Azad, F. Sabrina, S. Wasimi, Transformation of smart grid using ma- chine learning, in: 2019 29th Australasian Universities Power Engineer- ing Conference (AUPEC), 2019, pp. 1–6. doi:10.1109/AUPEC48547. 2019.211809

    2019

  16. [24]

    Z. Shi, W. Yao, Z. Li, L. Zeng, Y. Zhao, R. Zhang, Y. Tang, J. Wen, Ar- tificial intelligence techniques for stability analysis and control in smart grids: Methodologies, applications, challenges and future directions, Ap- plied Energy 278 (2020) 115733. doi:https://doi.org/10...

    2020

  17. [25]

    Aliyeva, N

    L. Aliyeva, N. Abdullayev, Hybrid deep learning approach towards smart grid stability prediction, in: 2024 IEEE 8th Energy Conference (ENERGYCON), 2024, pp. 1–5. doi:10.1109/ENERGYCON58629.2024. 10488774

    2024

  18. [26]

    A. K. Bashir, S. Khan, B. Prabadevi, N. Deepa, W. S. Alnumay, T. R. Gadekallu, P. K. R. Maddikunta, Comparative analysis of ma- chine learning algorithms for prediction of smart grid stability†, Interna- tional Transactions on Electrical Energy Systems 31 (9) (2021) e12706. do...

    2021

  19. [27]

    Gupta, G

    A. Gupta, G. Gurrala, P. S. Sastry, An online power system stability monitoring system using convolutional neural networks, IEEE Transac- tions on Power Systems 34 (2) (2019) 864–872. doi:10.1109/TPWRS. 2018.2872505

    2019

  20. [28]

    Z. Shi, W. Yao, L. Zeng, J. Wen, J. Fang, X. Ai, J. Wen, Convolutional neural network-based power system transient stability assessment and 33 instability mode prediction, Applied Energy 263 (2020) 114586. doi: https://doi.org/10.1016/j.apenergy.2020.114586

    2020

  21. [29]

    J. Tian, B. Wang, J. Li, Z. Wang, Adversarial attacks and defense for cnn based power quality recognition in smart grid, IEEE Transactions on Network Science and Engineering 9 (2) (2022) 807–819. doi:10. 1109/TNSE.2021.3135565

    2022

  22. [30]

    Aygul, M

    K. Aygul, M. Mohammadpourfard, M. Kesici, F. Kucuktezcan, I. Genc, Benchmark of machine learning algorithms on transient stability pre- diction in renewable rich power grids under cyber-attacks, Internet of Things 25 (2024) 101012. doi:https://doi.org/10.1016/j.iot. 2023.101012

    2024

  23. [31]

    Y. Chen, Y. Tan, D. Deka, Is machine learning in power systems vulnera- ble?, in: 2018 IEEE International Conference on Communications, Con- trol, and Computing Technologies for Smart Grids (SmartGridComm), 2018, pp. 1–6. doi:10.1109/SmartGridComm.2018.8587547

    2018

  24. [32]

    J. Tian, B. Wang, J. Li, C. Konstantinou, Adversarial attack and defense methods for neural network based state estimation in smart grid, IET Renewable Power Generation 16 (16) (2022) 3507–3518

    2022

  25. [33]

    Sayghe, J

    A. Sayghe, J. Zhao, C. Konstantinou, Evasion attacks with adversarial deep learning against power system state estimation, in: 2020 IEEE Power and Energy Society General Meeting (PESGM), 2020, pp. 1–5. doi:10.1109/PESGM41954.2020.9281719

    2020

  26. [34]

    Ahmadian, H

    S. Ahmadian, H. Malki, Z. Han, Cyber attacks on smart energy grids using generative adverserial networks, in: 2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP), 2018, pp. 942–946. doi:10.1109/GlobalSIP.2018.8646424

    2018

  27. [35]

    J. Li, Y. Yang, J. S. Sun, Searchfromfree: Adversarial measurements for machine learning-based energy theft detection, in: 2020 IEEE International Conference on Communications, Control, and Comput- ing Technologies for Smart Grids (SmartGridComm), 2020, pp. 1–6. doi:10.1109/Sm...

    2020

  28. [36]

    Mohsen, M

    S. Mohsen, M. Bajaj, H. Kotb, M. Pushkarna, S. Alphonse, S. S. Ghoneim, et al., Efficient artificial neural network for smart grid stabil- ity prediction, International Transactions on Electrical Energy Systems 2023 (2023)

    2023

  29. [37]

    Allal, H

    Z. Allal, H. N. Noura, O. Salman, K. Chahine, Leveraging the power of machine learning and data balancing techniques to evaluate stability in smart grids, Engineering Applications of Artificial Intelligence 133 (2024) 108304. doi:https://doi.org/10.1016/j.engappai.2024. 108304

    2024 doi

  30. [38]

    Mostafa, H

    N. Mostafa, H. S. M. Ramadan, O. Elfarouk, Renewable energy man- agement in smart grids by using big data analytics and machine learn- ing, Machine Learning with Applications 9 (2022) 100363. doi:https: //doi.org/10.1016/j.mlwa.2022.100363

    2022

  31. [39]

    J. E. Sullivan, D. Kamensky, How cyber-attacks in ukraine show the vulnerability of the us power grid, The Electricity Journal 30 (3) (2017) 30–35

    2017

  32. [40]

    Salazar, S

    L. Salazar, S. R. Castro, J. Lozano, K. Koneru, E. Zambon, B. Huang, R. Baldick, M. Krotofil, A. Rojas, A. A. Cardenas, A tale of two indus- troyers: It was the season of darkness, in: 2024 IEEE Symposium on Security and Privacy (SP), 2024, pp. 312–330. doi:10.1109/SP54263. 2024.00162

    2024

  33. [41]

    S. Xu, Y. Xia, H.-L. Shen, Analysis of malware-induced cyber attacks in cyber-physical power systems, IEEE Transactions on Circuits and Systems II: Express Briefs 67 (12) (2020) 3482–3486

    2020

  34. [42]

    I. Ali, A. I. A. Ahmed, A. Almogren, M. A. Raza, S. A. Shah, A. Khan, A. Gani, Systematic literature review on iot-based botnet attack, IEEE access 8 (2020) 212220–212232

    2020

  35. [43]

    O. Sen, D. van der Velde, P. Linnartz, I. Hacker, M. Henze, M. Andres, A. Ulbig, Investigating man-in-the-middle-based false data injection in a smart grid laboratory environment, in: 2021 IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), 2021, pp. 01–06. doi: ...

    2021

  36. [44]

    Krause, R

    T. Krause, R. Ernst, B. Klaer, I. Hacker, M. Henze, Cybersecurity in power grids: Challenges and opportunities, Sensors 21 (18) (2021) 6225

    2021

  37. [45]

    Inc., Crashoverride: Analysis of the threat to electric grid operations, accessed: 2025-03-12 (2017)

    D. Inc., Crashoverride: Analysis of the threat to electric grid operations, accessed: 2025-03-12 (2017). URL https://www.dragos.com/wp-content/uploads/ CrashOverride-01.pdf

    2017

  38. [46]

    J. C. Costa, T. Roxo, H. Proen¸ ca, P. R. M. In´ acio, How deep learning sees the world: A survey on adversarial attacks & defenses, IEEE Access 12 (2024) 61113–61136. doi:10.1109/ACCESS.2024.3395118

    2024

  39. [47]

    Tram` er, A

    F. Tram` er, A. Kurakin, N. Papernot, I. Goodfellow, D. Boneh, P. Mc- Daniel, Ensemble adversarial training: Attacks and defenses (2020). arXiv:1705.07204

    2020 arXiv

  40. [48]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y. Bengio, Generative adversarial networks, Communications of the ACM 63 (11) (2020) 139–144

    2020

  41. [49]

    Creswell, T

    A. Creswell, T. White, V. Dumoulin, K. Arulkumaran, B. Sengupta, A. A. Bharath, Generative adversarial networks: An overview, Vol. 35, 2018, pp. 53–65. doi:10.1109/MSP.2017.2765202

    2018

  42. [50]

    Alecci, M

    M. Alecci, M. Conti, F. Marchiori, L. Martinelli, L. Pajola, Your at- tack is too dumb: Formalizing attacker scenarios for adversarial trans- ferability, in: Proceedings of the 26th International Symposium on Research in Attacks, Intrusions and Defenses, RAID ’23, Association ...

    2023

  43. [51]

    Arzamasov, Electrical Grid Stability Simulated Data , UCI Machine Learning Repository, DOI: https://doi.org/10.24432/C5PG66 (2018)

    V. Arzamasov, Electrical Grid Stability Simulated Data , UCI Machine Learning Repository, DOI: https://doi.org/10.24432/C5PG66 (2018)

    2018 doi

  44. [52]

    S. M. Al-Selwi, M. F. Hassan, S. J. Abdulkadir, M. G. Ragab, A. Alqushaibi, E. H. Sumiea, Smart grid stability prediction using adap- tive aquila optimizer and ensemble stacked bilstm, Results in Engineer- ing 24 (2024) 103261. 36

    2024

  45. [53]

    Hangun, O

    B. Hangun, O. Eyecioglu, M. Beken, Forecasting the stability of a 4- node architecture smart grid using machine learning, in: 2022 10th In- ternational Conference on Smart Grid (icSmartGrid), IEEE, 2022, pp. 440–442

    2022

  46. [54]

    Hangun, O

    B. Hangun, O. Eyecioglu, O. Altun, Quantum computing approach to smart grid stability forecasting, in: 2024 12th International Conference on Smart Grid (icSmartGrid), IEEE, 2024, pp. 840–843

    2024

  47. [55]

    V. A. G. Raju, M. Mishra, J. G. Singh, J. Nayak, P. B. Dash, Smart grid stability prediction based on bayesian-optimised lgbm for smarter energy management, International Journal of Advanced Mechatronic Systems 11 (4) (2024) 226–241

    2024

  48. [56]

    M. K. Boutahir, A. Hessane, Y. Farhaoui, M. Azrour, An effective en- semble learning model to predict smart grid stability using genetic algo- rithms, in: Advanced Technology for Smart Environment and Energy, Springer, 2023, pp. 129–137

    2023

  49. [57]

    J. Mulo, P. Tian, A. Hussaini, H. Liang, W. Yu, Towards an adver- sarial machine learning framework in cyber-physical systems, in: 2023 IEEE/ACIS 21st International Conference on Software Engineering Re- search, Management and Applications (SERA), IEEE, 2023, pp. 138– 143

    2023

  50. [58]

    Sch¨ afer, C

    B. Sch¨ afer, C. Grabow, S. Auer, J. Kurths, D. Witthaut, M. Timme, Taming instabilities in power grid networks by decentralized control, The European Physical Journal Special Topics 225 (2016) 569–582

    2016

  51. [59]

    Arzamasov, P

    V. Arzamasov, P. Breviglieri, Augmented version of the original smart grid stability dataset, https://www.kaggle.com/datasets/ pcbreviglieri/smart-grid-stability/data , hosted on Kaggle (Jan- uary 2023)

    2023

  52. [60]

    Kim, Torchattacks: A pytorch repository for adversarial attacks, arXiv preprint arXiv:2010.01950 (2020)

    H. Kim, Torchattacks: A pytorch repository for adversarial attacks, arXiv preprint arXiv:2010.01950 (2020)

    2020 arXiv

  53. [61]

    Nicolae, M

    M.-I. Nicolae, M. Sinn, M. N. Tran, B. Buesser, A. Rawat, M. Wistuba, V. Zantedeschi, N. Baracaldo, B. Chen, H. Ludwig, et al., Adversarial robustness toolbox v1. 0.0, arXiv preprint arXiv:1807.01069 (2018). 37

    2018 arXiv

  54. [62]

    Ucar, A comprehensive analysis of smart grid stability prediction along with explainable artificial intelligence, Symmetry 15 (2) (2023)

    F. Ucar, A comprehensive analysis of smart grid stability prediction along with explainable artificial intelligence, Symmetry 15 (2) (2023). doi:10.3390/sym15020289

    2023 doi

  55. [63]

    Raspberry Pi Foundation, Raspberry Pi: Putting the power of com- puting and digital making into the hands of people all over the world, https://www.raspberrypi.org/ (February 2023). 38

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.