pith. sign in

arxiv: 2509.22795 · v1 · pith:FD5XL463new · submitted 2025-09-26 · 📡 eess.SP · cs.AI· cs.SY· eess.SY

Generative Modeling and Decision Fusion for Unknown Event Detection and Classification Using Synchrophasor Data

Yi Hu , Zheyuan Cheng This is my paper

Pith reviewed 2026-05-22 13:23 UTC · model grok-4.3

classification 📡 eess.SP cs.AIcs.SYeess.SY
keywords synchrophasor dataevent detectionVAE-GANunknown event classificationpower system monitoringanomaly detectiondecision fusion
0
0 comments X

The pith

A VAE-GAN trained only on normal conditions detects known power events and assigns unseen disturbances to a new category.

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

The paper presents a framework that overcomes the inability of supervised classifiers to handle rare or previously unseen disturbances in power grids. It trains a variational autoencoder-generative adversarial network exclusively on normal operating data to extract reconstruction error and discriminator error as indicators of anomalies. These indicators are processed through sliding windows to form spatiotemporal matrices across multiple phasor measurement units, then fused using either a simple threshold rule or a more robust convex hull approach. The resulting system identifies documented events while routing novel disturbances into an explicit unknown class. Experimental tests show higher accuracy than standard machine learning, deep learning, and envelope-based methods on synchrophasor streams.

Core claim

By modeling normal grid behavior with a VAE-GAN and deriving anomaly scores from both reconstruction and discriminator outputs, the framework builds sliding-window spatiotemporal matrices and applies decision fusion across PMUs to classify known events accurately while systematically placing previously unseen disturbances into a dedicated new category.

What carries the argument

VAE-GAN trained solely on normal operating conditions to generate reconstruction error and discriminator error as anomaly indicators, combined with sliding-window matrix construction and threshold or convex-hull decision fusion.

If this is right

  • The method identifies known events while routing unseen disturbances into an explicit new category.
  • It achieves higher accuracy than machine learning, deep learning, and envelope-based baselines.
  • The convex-hull decision strategy provides robustness when error distributions are complex.
  • Spatiotemporal fusion across PMUs supports wide-area monitoring of power system events.

Where Pith is reading between the lines

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

  • The same generative-model-plus-fusion structure could be tested on other streaming sensor networks for anomaly handling.
  • Replacing the convex-hull step with a faster approximation might preserve robustness while reducing latency for real-time use.
  • Collecting a small set of post-deployment unknown events could allow the model to be updated without full retraining.

Load-bearing premise

The reconstruction and discriminator errors produced by a VAE-GAN trained only on normal conditions will reliably separate both known and previously unseen disturbances in real synchrophasor data.

What would settle it

Apply the trained model to a labeled synchrophasor dataset containing both documented events and deliberately introduced novel disturbances, then measure whether the novel cases are consistently routed to the unknown class without being mislabeled as known events or normal operation.

Figures

Figures reproduced from arXiv: 2509.22795 by Yi Hu, Zheyuan Cheng.

Figure 1
Figure 1. Figure 1: Event classification accuracy degradation due to unseen event types. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Generative AI-based decision fusion framework. B. Signal Selection and Data Pre-processing In the synchrophasor data, there are many signals that are available for event detection and classification. After extensive testing with field recorded synchrophasor data, the following signals provide the best detection and classification performance: Voltage magnitude at three phases 𝑽𝒂, 𝑽𝒃, and 𝑽𝒄 , Positive sequ… view at source ↗
Figure 3
Figure 3. Figure 3: VAE-GAN model architecture. Under normal operating conditions, the model learns to accurately reconstruct time-series data from phasor measurement units (PMUs), resulting in consistently low reconstruction and discrimination errors. However, when an abnormal event occurs, the input pattern deviates from the learned distribution of normal behavior. As a result, the model’s reconstruction becomes less accura… view at source ↗
Figure 4
Figure 4. Figure 4: 2D error feature space illustration. Method 1: Threshold based decision. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The sliding window mechanism [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Flowchart for event detection and classification identification. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Event detection based on different decision [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: demonstrates the ability of the proposed VAE￾GAN framework to detect and localize a power system event using 1-minute phasor measurement unit (PMU) data. The left subplot shows normalized voltage magnitude signals from multiple PMU channels over a 60-second interval. While most signals remain steady, a sharp drop occurs around timestamp 01:09:36, clearly indicating the presence of an event. A smaller distu… view at source ↗
Figure 9
Figure 9. Figure 9: Example events detected by only one method. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Example event detected by both methods with different durations. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: presents an example of the detection and classification outcomes for a disturbance event. The left panel shows the voltage magnitude measurement, where a sharp deviation indicates the presence of an event. On the right, the detection matrix (top) provides a binary representation in which black squares correspond to abnormal segments flagged by the detection pathway. These blocks appear prominently in the … view at source ↗
read the original abstract

Reliable detection and classification of power system events are critical for maintaining grid stability and situational awareness. Existing approaches often depend on limited labeled datasets, which restricts their ability to generalize to rare or unseen disturbances. This paper proposes a novel framework that integrates generative modeling, sliding-window temporal processing, and decision fusion to achieve robust event detection and classification using synchrophasor data. A variational autoencoder-generative adversarial network is employed to model normal operating conditions, where both reconstruction error and discriminator error are extracted as anomaly indicators. Two complementary decision strategies are developed: a threshold-based rule for computational efficiency and a convex hull-based method for robustness under complex error distributions. These features are organized into spatiotemporal detection and classification matrices through a sliding-window mechanism, and an identification and decision fusion stage integrates the outputs across PMUs. This design enables the framework to identify known events while systematically classifying previously unseen disturbances into a new category, addressing a key limitation of supervised classifiers. Experimental results demonstrate state-of-the-art accuracy, surpassing machine learning, deep learning, and envelope-based baselines. The ability to recognize unknown events further highlights the adaptability and practical value of the proposed approach for wide-area event analysis in modern power systems.

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 / 2 minor

Summary. The manuscript proposes a framework for detecting and classifying power system events from synchrophasor data. A VAE-GAN is trained exclusively on normal operating conditions to extract reconstruction error and discriminator error as anomaly indicators. These indicators are arranged into spatiotemporal matrices via a sliding-window approach and fed to either a threshold-based rule or a convex-hull rule; the resulting per-PMU decisions are fused to label known events while routing all unseen disturbances into a single new category. The paper claims this yields state-of-the-art accuracy that surpasses machine-learning, deep-learning, and envelope-based baselines.

Significance. If the error distributions prove separable, the approach would address a genuine limitation of purely supervised classifiers by enabling systematic flagging of novel disturbances without retraining. The integration of generative modeling with sliding-window spatiotemporal processing and PMU-level fusion is a coherent design choice that could be practically useful for wide-area situational awareness.

major comments (2)
  1. Section 3.2 (Anomaly Indicators and Decision Strategies): The central claim that reconstruction-plus-discriminator errors will reliably separate known-event classes from the unknown class (and from each other) is load-bearing, yet the manuscript provides no quantitative analysis of per-class distribution overlap, no Kolmogorov-Smirnov statistics, and no ablation that removes the convex-hull step. Without such evidence the fusion stage cannot be guaranteed to produce the claimed per-event labels while still isolating unknowns.
  2. Section 4 (Experimental Results): The abstract asserts state-of-the-art accuracy and superiority over baselines, but the reported results lack error bars, dataset size and composition details, cross-validation protocol, and ablation tables that isolate the contribution of the convex-hull versus threshold rule. These omissions prevent verification that the performance gain is not due to post-hoc threshold tuning or baseline implementation choices.
minor comments (2)
  1. Notation for the spatiotemporal detection matrix (Eq. (7) or equivalent) is introduced without an accompanying numerical example; adding a small illustrative matrix would improve readability.
  2. The description of the sliding-window length and overlap parameters appears only in the text; tabulating the exact values used for each experiment would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The comments identify areas where additional quantitative support and experimental rigor will strengthen the manuscript. We address each major comment below and commit to incorporating the suggested improvements in the revised version.

read point-by-point responses

  1. Referee: Section 3.2 (Anomaly Indicators and Decision Strategies): The central claim that reconstruction-plus-discriminator errors will reliably separate known-event classes from the unknown class (and from each other) is load-bearing, yet the manuscript provides no quantitative analysis of per-class distribution overlap, no Kolmogorov-Smirnov statistics, and no ablation that removes the convex-hull step. Without such evidence the fusion stage cannot be guaranteed to produce the claimed per-event labels while still isolating unknowns.

    Authors: We agree that explicit quantitative evidence of separability is required to substantiate the central claim. In the revised manuscript we will add Kolmogorov-Smirnov tests comparing the reconstruction and discriminator error distributions across known-event classes and the unknown class, together with overlap metrics such as the Bhattacharyya coefficient. We will also include an ablation that removes the convex-hull decision rule and reports performance using only the threshold-based rule. These additions will be placed in Section 3.2 and will directly support the reliability of the subsequent fusion stage. revision: yes

  2. Referee: Section 4 (Experimental Results): The abstract asserts state-of-the-art accuracy and superiority over baselines, but the reported results lack error bars, dataset size and composition details, cross-validation protocol, and ablation tables that isolate the contribution of the convex-hull versus threshold rule. These omissions prevent verification that the performance gain is not due to post-hoc threshold tuning or baseline implementation choices.

    Authors: We acknowledge that the current experimental section lacks several elements necessary for full reproducibility and verification. The revised manuscript will report error bars (standard deviation across repeated runs with different random seeds), provide complete dataset size, composition, and event-type details, describe the cross-validation protocol, and include ablation tables that isolate the contribution of the convex-hull rule versus the threshold rule. We will also clarify the implementation of all baselines to ensure the reported gains are attributable to the proposed components rather than implementation or tuning differences. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes a VAE-GAN trained exclusively on normal operating conditions to extract reconstruction and discriminator errors as anomaly indicators, which are then processed via sliding-window spatiotemporal matrices, threshold or convex-hull decision rules, and PMU-level fusion to label known events while routing unseen disturbances to a new category. This chain relies on standard generative modeling practices plus empirically chosen decision heuristics rather than any self-definitional reduction, fitted parameter renamed as prediction, or load-bearing self-citation. No equations or steps in the provided description reduce the claimed detection/classification performance to the inputs by construction; the framework is presented as an independent application whose validity rests on experimental results against baselines.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework depends on the domain assumption that normal grid behavior is sufficiently stationary to be captured by a single generative model and that error signals from that model generalize to unseen events; no free parameters are explicitly named in the abstract, and no new physical entities are introduced.

free parameters (1)
  • anomaly decision threshold
    Used in the threshold-based rule; must be chosen or tuned on data to balance detection and false alarms.
axioms (1)
  • domain assumption Reconstruction error and discriminator error from a VAE-GAN trained on normal data serve as effective anomaly indicators for both known and unknown power system events.
    Invoked when the paper extracts these errors as features for the detection and classification matrices.

pith-pipeline@v0.9.0 · 5749 in / 1438 out tokens · 33497 ms · 2026-05-22T13:23:30.570702+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Statistical feature extraction is used to classify events from PMU data stream [8]

    or singular value decomposition (SVD) [7] to capture spatial correlations and structural changes. Statistical feature extraction is used to classify events from PMU data stream [8]. A dynamic programming -based swinging door trending (DPSDT) method is proposed for high -precision PMU event detection [9]. While effective in some settings, these rule - base...

    work page

  2. [2]

    Such constraints often result in poor generalization of event classifiers when applied to unseen conditions

    trains an event classifier on only 57 labeled line events, and [11] utilizes several hundred labeled frequency events W 2 from the FNET/GridEye system to develop a CNN -based detector. Such constraints often result in poor generalization of event classifiers when applied to unseen conditions. To mitigate data scarcity, recent works have explored the use o...

    work page

  3. [3]

    The encoder network receives normalized PMU measurements at a 5 -second length 𝑴

    Model Configuration As illustrated in Figure 3, the system consists of three major components: Encoder, Decoder/Generator, and Discriminator. The encoder network receives normalized PMU measurements at a 5 -second length 𝑴. It outputs two vectors: mean vector 𝝁, and standard deviation vector 𝝈. A latent vector 𝒛 is then sampled using the reparameterizatio...

    work page

  4. [4]

    Loss Function Designation for VAE-GAN The total loss for training the VAE-GAN model is: ℒ𝑉𝐴𝐸−𝐺𝐴𝑁 = ℒ𝑟𝑒𝑐𝑜𝑛 + 𝜆1ℒ𝐾𝐿 + 𝜆2ℒ𝑎𝑑𝑣 (5) where 𝜆1 and 𝜆2 are scaling coefficients. Reconstruction Loss that measures the difference between original data 𝑴 and reconstructed data 𝑴̂ 𝑖 using a hybrid of MSE and a max - penalty term: ℒ𝑟𝑒𝑐𝑜𝑛 = 1 𝑁 ∑ ‖𝑴𝑖 − 𝑴̂ 𝑖‖ 2𝑁 𝑖=1 + max...

    work page

  5. [5]

    We develop two complementary decision-making strategies: Fig

    Decision Metrics After extracting the reconstruction error 𝑒𝑟𝑒𝑐𝑜𝑛 and discriminator output error 𝑒𝐷 from the VAE –GAN model, these two features are mapped to a 2D error feature space to determine whether the measured data corresponds to normal operation condition or an event condition. We develop two complementary decision-making strategies: Fig. 4. 2D er...

    work page

  6. [6]

    Event Detection and Classification Identification Figure 6 illustrates the identification process for event detection and classification based on the spatiotemporal matrices generated in the sliding windowing stage. For each PMU row in the matrix, the algorithm scans the sequence of outputs to identify the maximum number of consecutive occurrences of the ...

    work page

  7. [7]

    Decision Fusion The decision fusion process integrates the outputs of the detection module and the classification module to produce the final event decision. As shown in Table I, the detection output provides a binary indication of whether an event is present, while the classification module predicts an event type if the data segment deviates from normal ...

    work page

  8. [8]

    The scalability of the algorithm is tested on a laptop PC

    Event Detection Speed High speed event detection is the key when dealing with massive amount of synchrophasor data. The scalability of the algorithm is tested on a laptop PC. It only takes around 5 seconds for the proposed detection to scan through a 1 -minute synchrophasor dataset and create a list of events. It shows that the proposed detection method i...

    work page

  9. [9]

    Figure 7 illustrates the distribution of individual PMU data segments in the 2D feature space defined by the reconstruction error and discriminator error

    Event Detection on 5-second Window for Single PMU To validate the effectiveness of the proposed VAE -GAN- based event detection framework, a series of detection experiments were conducted using labeled phasor measurement unit (PMU) time -series data containing both normal and abnormal system conditions. Figure 7 illustrates the distribution of individual ...

    work page

  10. [10]

    The left subplot shows normalized voltage magnitude signals from multiple PMU channels over a 60 -second interval

    Event Detection on 1-minute Window for Multiple PMUs Figure 8 demonstrates the ability of the proposed VAE - GAN framework to detect and localize a power system event using 1-minute phasor measurement unit (PMU) data. The left subplot shows normalized voltage magnitude signals from multiple PMU channels over a 60 -second interval. While most signals remai...

    work page

  11. [11]

    The left panel shows the voltage magnitude measurement, where a sharp deviation indicates the presence of an event

    2-D Matrices Generation Figure 11 presents an example of the detection and classification outcomes for a disturbance event. The left panel shows the voltage magnitude measurement, where a sharp deviation indicates the presence of an event. On the right, the detection matrix (top) provides a binary representation in which black squares correspond to abnorm...

    work page

  12. [12]

    Event Identification and Decision Fusion The binary detection matrix and non -binary classification matrix introduced earlier provide the foundational inputs for the next stage of event decision -making. While these matrices clearly capture when anomalies occur and how they may be categorized, further processing is required to transform them into reliable...

    work page

  13. [13]

    Without the sliding window (top row), the classification accuracy is 81.89%

    Event Classification for Unknown Event Types Table IV demonstrates the effectiveness of the proposed method in improving event classification, particularly in the presence of unknown event types. Without the sliding window (top row), the classification accuracy is 81.89%. Accuracy means the ratio of correctly classified samples and total number of samples...

    work page

  14. [14]

    Ablation Study To better understand the contribution of the decision fusion and sliding window mechanisms in the proposed framework, an ablation study was performed by evaluating four different scenarios, as shown in Table VI. TABLE VI Event Classification Performance in Different Scenarios Decision Fusion Sliding Window Detection Accuracy Classification ...

    work page

  15. [15]

    The 2003 blackout: Solutions that won’t cost a fortune,

    D. White, A. Roschelle, P. Peterson, D. Schlissel, B. Biewald, and W. Steinhurst, “The 2003 blackout: Solutions that won’t cost a fortune,” Electricity J., vol. 16, no. 9, pp. 43–53, 2003

    work page 2003

  16. [16]

    Ice Storm Makes Christmas a Dark Day for Tens of Thousands

    “Ice Storm Makes Christmas a Dark Day for Tens of Thousands.” Accessed: Set. 08, 2025 . [Online]. Available: https://www.cbc.ca/news/canada/ice-stormmeansdark-christmas-for- tens-of-thousands-1.2476164

    work page 2025

  17. [17]

    Wavelet - based event detection method using PMU data,

    D.-I. Kim, T. Y. Chun, S. -H. Yoon, G. Lee, and Y. -J. Shin, “Wavelet - based event detection method using PMU data,” IEEE Transactions on Smart Grid, vol. 8, no. 3, pp. 1154–1162, Oct. 2015

    work page 2015

  18. [18]

    Event detection method for the PMUs synchrophasor data,

    S.-W. Sohn, A. J. Allen, S. Kulkarni, W. M. Grady, and S. Santoso, “Event detection method for the PMUs synchrophasor data,” in 2012 IEEE Power Electronics and Machines in Wind Applications, Jul. 2012, pp. 1–7

    work page 2012

  19. [19]

    Dimensionality reduction of synchrophasor data for early event detection: Linearized analysis,

    L. Xie, Y. Chen, and P. Kumar, “Dimensionality reduction of synchrophasor data for early event detection: Linearized analysis,” IEEE Transactions on Power Systems, vol. 29, no. 6, pp. 2784 –2794, Nov. 2014

    work page 2014

  20. [20]

    Real time anomaly detection in wide area monitoring of smart grids,

    J. Wu, J. Xiong, P. Shil, and Y. Shi, “Real time anomaly detection in wide area monitoring of smart grids,” in 2014 IEEE/ACM International Conference on Computer-Aided Design (ICCAD), Nov. 2014, pp. 197 – 204

    work page 2014

  21. [21]

    Real -time event identification through low-dimensional subspace characterization of high -dimensional synchrophasor data,

    W. Li, M. Wang, and J. H. Chow, “Real -time event identification through low-dimensional subspace characterization of high -dimensional synchrophasor data,” IEEE Trans. Power Syst., vol. 33, no. 5, pp. 4937 – 4947, Sep. 2018

    work page 2018

  22. [22]

    Event detection and its signal characterization in PMU data stream,

    S. S. Negi, N. Kishor, K. Uhlen, and R. Negi, “Event detection and its signal characterization in PMU data stream,” IEEE Trans. Ind. Informat., vol. 13, no. 6, pp. 3108–3118, Dec. 2017

    work page 2017

  23. [23]

    A novel event detection method using PMU data with high precision,

    M. Cui, J. Wang, J. Tan, A. R. Florita, and Y. Zhang, “A novel event detection method using PMU data with high precision,” IEEE Trans. Power Syst., vol. 34, no. 1, pp. 454–466, Jan. 2019

    work page 2019

  24. [24]

    Preliminary work to classify the disturbance events recorded by phasor measurement units,

    O. P. Dahal and S. M. Brahma, “Preliminary work to classify the disturbance events recorded by phasor measurement units,” in Proc. IEEE Power Energy Soc. Gen. Meeting, 2012, pp. 1–8

    work page 2012

  25. [25]

    Frequency disturbance event detection based on synchrophasors and deep learning,

    W. Wang et al., “Frequency disturbance event detection based on synchrophasors and deep learning,” IEEE Trans. Smart Grid, vol. 11, no. 4, pp. 3593–3605, Jul. 2020

    work page 2020

  26. [26]

    PMU-data-driven event classification in power transmission grids,

    I. Niazazari, Y . Liu, A. Ghasenikhani, S. Biswas, H. Livani, L. Yang, and V . A. Centeno , “PMU-data-driven event classification in power transmission grids,” in Proc. IEEE Power Energy Soc. Innov. Smart Grid Technol. Conf. (ISGT), 2021, pp. 1–5

    work page 2021

  27. [27]

    Hierarchical convolutional neural networks for event classification on PMU measurements,

    M. Pavlovski, M. Alqudah, T. Dokic, A. A. Hai, M. Kezunovic, and Z. Obradovic, “Hierarchical convolutional neural networks for event classification on PMU measurements,” IEEE Trans. Instrum. Meas., vol. 70, pp. 1–13, 2021

    work page 2021

  28. [28]

    Learningbased real-time event identification using rich real PMU data,

    Y. Yuan, Y. Guo, K. Dehghanpour, Z. Wang, and Y. Wang, “Learningbased real-time event identification using rich real PMU data,” IEEE Trans. Power Syst., vol. 36, no. 6, pp. 5044–5055, Nov. 2021

    work page 2021

  29. [29]

    Big data processing for power grid event detection,

    B. P. Leao, D. Fradkin, Y. Wang, and S. Suresh, “Big data processing for power grid event detection,” in Proc. IEEE Int. Conf. Big Data (Big Data), 2020, pp. 4128–4136

    work page 2020

  30. [30]

    Deep-Learning based Multiple Class Events Detection and Classification using Micro -PMU Data

    R. Chandrakar, R. Dubey, and B. Panigrahi. “Deep-Learning based Multiple Class Events Detection and Classification using Micro -PMU Data.” In 2024 8th International Conference on Green Energy and Applications (ICGEA), pp. 137-142. IEEE, 2024

    work page 2024

  31. [31]

    Online power system event detection via bidirectional generative adversarial networks

    Y. Cheng, N. Yu, B. Foggo, and K. Yamashita. “Online power system event detection via bidirectional generative adversarial networks. ” IEEE Transactions on Power Systems 37, no. 6 (2022): 4807-4818

    work page 2022

  32. [32]

    Smart grid line event classification using supervised learning over PMU data streams,

    D. Nguyen, R. Barella, S. A. Wallace, X. Zhao, and X. Liang, “Smart grid line event classification using supervised learning over PMU data streams,” in Proc. 6th Int. Green Sustain. Comput. Conf. (IGSC), 2015, pp. 1–8

    work page 2015

  33. [33]

    Data- driven event detection of power systems based on unequal-interval reduction of PMU data and local outlier factor,

    S. Liu, Y. Zhao, Z. Lin, Y. Liu, Y. Ding, L . Yang, and S . Yi, “Data- driven event detection of power systems based on unequal-interval reduction of PMU data and local outlier factor,” IEEE Trans. Smart Grid, vol. 11, no. 2, pp. 1630–1643, Mar. 2020

    work page 2020

  34. [34]

    Generative adversarial networks-based synthetic PMU data creation for improved event classification,

    X. Zheng, B. Wang, D. Kalathil, and L. Xie, “Generative adversarial networks-based synthetic PMU data creation for improved event classification,” IEEE Open Access J. Power Energy, vol. 8, pp. 68 –76, 2021

    work page 2021

  35. [35]

    Automated High-Speed Power System Event Detection and Classification Using Synchrophasor Data

    Z. Cheng, J . Schmidt, Y . Hu, D . Hart, and J . Holbach. “Automated High-Speed Power System Event Detection and Classification Using Synchrophasor Data.” U.S. Patent Application 18/131,155, filed October 10, 2024

    work page 2024