Partial Discharge Detection on Aerial Covered Conductors Using Time-Series Decomposition and Long Short-term Memory Network

Jessie Sun; Ming Dong

arxiv: 1907.03378 · v1 · pith:7GFOXXT2new · submitted 2019-07-08 · 📡 eess.SP

Partial Discharge Detection on Aerial Covered Conductors Using Time-Series Decomposition and Long Short-term Memory Network

Ming Dong , Jessie Sun This is my paper

Pith reviewed 2026-05-25 01:25 UTC · model grok-4.3

classification 📡 eess.SP

keywords partial discharge detectioncovered conductorsLSTM networktime-series decompositionvoltage waveform classificationENET datasetfault detectionsignal processing

0 comments

The pith

Time-series decomposition and LSTM networks detect partial discharges on aerial covered conductors more effectively than standard methods.

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

Aerial covered conductors provide higher reliability and wildlife protection but their faults often produce only partial discharges rather than the overcurrent signals that standard protection devices catch. The paper develops a classification approach that decomposes voltage time-series signals recorded by ENET stray-field meters and routes the results through an LSTM network after specialized feature engineering. When evaluated on the public ENET dataset released on Kaggle, the method outperforms several traditional classification techniques. A reader would care because reliable detection could allow utilities to address tree-contact and phase-to-ground events on these conductors without replacing existing infrastructure.

Core claim

The authors state that their method, built on time-series decomposition of voltage waveforms together with an LSTM network and custom feature engineering, recognizes partial-discharge activities on covered conductors with superior performance and practicality when tested against conventional classifiers on the ENET public dataset.

What carries the argument

Time-series decomposition combined with a Long Short-Term Memory network that classifies voltage waveforms captured from the electric stray field along covered conductors.

If this is right

The method identifies tree-branch and ground-fault events that produce partial discharges without triggering overcurrent protection.
It achieves higher accuracy than traditional classifiers on the publicly released ENET dataset.
The feature-engineering step improves the LSTM's ability to extract useful patterns from the decomposed signals.
The approach operates directly on meter data already collected, supporting practical field use.

Where Pith is reading between the lines

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

Utilities could integrate the technique into existing covered-conductor monitoring networks if the patterns prove stable across additional sites.
The same decomposition-plus-LSTM pipeline might be tested on other power-system signals such as current or acoustic data for related fault types.
An edge-device version running on streaming meter output could be evaluated to measure detection latency in operational settings.

Load-bearing premise

The voltage waveforms recorded by the ENET meter contain repeatable, learnable patterns that distinguish partial-discharge events from normal operation across the range of real-world conditions the meter will encounter.

What would settle it

A new collection of voltage waveforms from different weather, load, or conductor conditions on which the LSTM model systematically misclassifies partial-discharge events or fails to flag them would falsify the performance claim.

Figures

Figures reproduced from arXiv: 1907.03378 by Jessie Sun, Ming Dong.

read the original abstract

Nowadays, aerial covered conductors (CC) are increasingly used in many places of the world due to their higher operational reliability, reduced construction space and effective protection for wildlife animals. In spite of these advantages, a major challenge of using CC is that the ordinary upstream protection devices are not able to detect the phase-to-ground faults and the frequent tree/tree branch hitting conductor events on such conductors. This is because these events only lead to partial discharge (PD) activities rather than overcurrent signatures typically seen on bare conductors. To solve this problem, in recent years, ENET Center in Czech Republic (ENET) devised a simple meter to measure the voltage signal of the electric stray field along CC, aiming to detect the above hazardous PD activities. In 2018, ENET shared a large amount of waveform data recorded by their meter on Kaggle, the world's largest data science collaboration platform, encouraging worldwide experts to develop an effective pattern recognition method for the acquired signals. In response, we developed a unique method based on time-series decomposition and Long Short-Term Memory Network (LSTM) in addition to unique feature engineering process to recognize PD activities on CC. The proposed method is tested on the ENET public dataset and compared to various traditional classification methods. It demonstrated superior performance and great practicality.

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.

Desk Editor's Note private letter to a colleague

Applies known decomposition and LSTM methods to the ENET PD dataset for covered conductors but reports no metrics or validation details to support the superiority claim.

read the letter

The paper takes time-series decomposition plus LSTM, adds feature engineering, and runs it on the ENET Kaggle waveforms to detect partial discharges on aerial covered conductors. It reports that this beats various traditional classifiers and calls the result practical for real distribution networks. That is the main contribution: a new target domain rather than a new algorithm. The problem itself is real—covered conductors hide tree-contact faults from standard overcurrent protection—and the choice to work with the public ENET release is sensible for benchmarking. The pipeline description is clear enough that someone could reproduce the steps. The framing of the operational gap in medium-voltage lines is also straightforward and useful. The evaluation side is thin. The abstract asserts superior performance without any accuracy numbers, dataset size, class balance, cross-validation method, or ablation on the decomposition or LSTM components. That absence makes the central claim impossible to check from the given text. The single-dataset setup also leaves the practicality statement unsupported; nothing is said about splits by season, weather, conductor age, or PD source type, so the stress-test concern about narrow distribution holds. A reader working on ML for power-system fault detection might still find the pipeline idea worth trying on their own data. The work shows clear engagement with the application but does not yet supply the evidence needed for a referee to judge whether the method actually generalizes or improves on existing practice. I would not send it for peer review until the quantitative results and some form of external validation are added.

Referee Report

3 major / 2 minor

Summary. The paper proposes a PD detection pipeline for aerial covered conductors that combines time-series decomposition, custom feature engineering, and an LSTM classifier. It evaluates the method on the public ENET Kaggle dataset and asserts that the approach outperforms several traditional classifiers while offering practical value for real-world monitoring.

Significance. If the performance claims are substantiated with proper metrics and generalization evidence, the work could address a genuine gap in protection for covered-conductor lines where conventional overcurrent devices fail. The decision to release and use a public dataset is a positive step toward reproducibility in this application domain.

major comments (3)

[Abstract, Experimental Results] Abstract and Experimental Results section: the claim of 'superior performance' is stated without any numerical metrics (accuracy, F1, AUC, etc.), dataset cardinality, class balance, or cross-validation protocol. This absence prevents verification of the central empirical claim.
[Method, Evaluation] Method and Evaluation sections: no ablation study isolates the contribution of the time-series decomposition step versus the LSTM or the feature-engineering pipeline alone, leaving the necessity of the proposed combination untested.
[Evaluation] Evaluation section: the manuscript gives no indication of stratified splits by season, weather, conductor age, or PD source type, nor any held-out external validation set. This directly weakens the 'great practicality' assertion when the ENET release may represent a narrow distribution of events.

minor comments (2)

[Method] Notation for the decomposition components and LSTM hyperparameters is introduced without a consolidated table, making reproduction more difficult.
[Evaluation] The comparison baselines are described only generically ('various traditional classification methods'); explicit listing of the algorithms and their hyperparameter settings would strengthen the comparison.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and indicate the revisions we will make to strengthen the paper.

read point-by-point responses

Referee: [Abstract, Experimental Results] Abstract and Experimental Results section: the claim of 'superior performance' is stated without any numerical metrics (accuracy, F1, AUC, etc.), dataset cardinality, class balance, or cross-validation protocol. This absence prevents verification of the central empirical claim.

Authors: We agree that the abstract does not contain numerical performance values. We will revise the abstract to include the key metrics (accuracy, F1-score, AUC) achieved by the proposed method on the ENET dataset. The Experimental Results section will be updated to explicitly report dataset cardinality, class balance, and the cross-validation protocol (including the number of folds). revision: yes
Referee: [Method, Evaluation] Method and Evaluation sections: no ablation study isolates the contribution of the time-series decomposition step versus the LSTM or the feature-engineering pipeline alone, leaving the necessity of the proposed combination untested.

Authors: We acknowledge that the current manuscript does not contain an ablation study. We will add an ablation study in the revised version that systematically removes or replaces each component (time-series decomposition, feature engineering, and LSTM) to quantify their individual contributions to overall performance. revision: yes
Referee: [Evaluation] Evaluation section: the manuscript gives no indication of stratified splits by season, weather, conductor age, or PD source type, nor any held-out external validation set. This directly weakens the 'great practicality' assertion when the ENET release may represent a narrow distribution of events.

Authors: The public ENET Kaggle dataset does not provide metadata labels for season, weather, conductor age, or PD source type, so stratified splits on these attributes were not feasible. We will add an explicit statement of this limitation and a discussion of its implications for generalization. No external held-out set beyond the public release was available; we will note this constraint and qualify the practicality claims accordingly. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical ML pipeline on external public dataset

full rationale

The paper presents a supervised classification pipeline (time-series decomposition + LSTM + feature engineering) trained and tested on the external ENET public dataset released on Kaggle. No equations, derivations, or 'predictions' are defined inside the paper that reduce reported accuracy or practicality claims to fitted parameters or self-referential constructions. Performance is measured by direct comparison against traditional classifiers on the same held-out data; no self-citation chains, uniqueness theorems, or ansatzes are invoked as load-bearing steps. This is a standard empirical ML application with no load-bearing internal reductions.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; the central claim rests on the unstated premise that the Kaggle waveforms are representative of field conditions and that LSTM can extract generalizable PD signatures from them.

pith-pipeline@v0.9.0 · 5757 in / 1073 out tokens · 22102 ms · 2026-05-25T01:25:29.048160+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 6 canonical work pages

[1]

Covered conductor is reliable solution for electricity distribution,

M. Bruun, “Covered conductor is reliable solution for electricity distribution,” 2011, Online : www.ensto.com

2011
[2]

An improved technique for online PD detection on covered conductor lines,

W. Zhang, Z. Hou, H. Li, C. Liu and N. Ma, “An improved technique for online PD detection on covered conductor lines,” IEEE Trans. on Power Delivery, vol. 29, no. 2, pp. 972-973, April 2014

2014
[3]

Modeling and experimental verification of on -line PD detection in MV covered-conductor Aerial networks,

G. M. Hashmi, M. Lehtonen and M. Nordman, “Modeling and experimental verification of on -line PD detection in MV covered-conductor Aerial networks,” IEEE Trans. Dielectr. Electr. Insul., vol. 17, no.1, pp. 167–180, Feb. 2010

2010
[4]

ENET Power Line Fault Detection ,

ENET Centre , “ ENET Power Line Fault Detection ,” 2018, Online: http://cenet.vsb.cz/en/informace/enet-centre/160/enet-centre-introduction.html 20

2018
[5]

Available: https://www.kaggle.com/c/vsb-power-line-fault-detection
[6]

Available: https://parquet.apache.org/documentation/latest/
[7]

STL: A Seasonal -Trend Decomposition,

R. B. Cleveland, W. S. Cleveland, J. E. McRae and I. Terpenning, “STL: A Seasonal -Trend Decomposition,” Journal of Official Statistics, 6(1), pp.3-73, 1990

1990
[8]

Sequence learning: from recognition and prediction to sequential decision making,

R. Sun and C. L. Giles, “Sequence learning: from recognition and prediction to sequential decision making,” IEEE Intelligent Systems, vol. 16, no. 4, pp. 67- 70, July-Aug. 2001

2001
[9]

Forecasting the volatility of s tock price index: A hybrid model integrating LSTM with multiple GARCH -type models ,

H. Y. Kim and C. H. Won, “Forecasting the volatility of s tock price index: A hybrid model integrating LSTM with multiple GARCH -type models ,” Expert Systems with Applications , vol. 103, pp. 25 -37, Aug. 2018

2018
[10]

Hourly day -ahead solar irradiance prediction u sing weather forecasts by LSTM ,

X. Qing and Y. Niu, “Hourly day -ahead solar irradiance prediction u sing weather forecasts by LSTM ,” Energy, vol. 148, pp. 461-468, Apr. 2018

2018
[11]

Machine translation using deep learning: an overview,

S. P. Singh, A. Kumar, H. Darbari, L. Singh, A. Rastogi and S. Jain, “Machine translation using deep learning: an overview,” 2017 International Conference on Computer, Communications and Electronics (Comptelix), pp. 162-167, 2017

work page 2017
[12]

Learning long -term dependencies with gradient descent is difficult,

Y. Bengio, P. Simard and P. Frasconi, “Learning long -term dependencies with gradient descent is difficult, ” IEEE Trans. on Neural Networks, vol. 5, pp. 157-166, 1994

1994
[13]

Gradient flow in recurrent nets: the difficulty of learning long -term dependencies,

J. F. Kolen and S. C. Kremer, “Gradient flow in recurrent nets: the difficulty of learning long -term dependencies,” A Field Guide to Dynamical Recurrent Networks , IEEE, 2001

2001
[14]

Long short -term memory,

S. Hochreiter and J. Schmidhuber, “Long short -term memory,” Neural computation , vol.9, pp.1735 - 1780, 1997

1997
[15]

Sequence to sequence learning with neural networks,

I. Sutskever, O. Vinyals and Q. V. Le, “Sequence to sequence learning with neural networks,” Advances in Neural Information Processing Systems, pp. 3104-3112, 2014

work page 2014
[16]

R. S. Michalski, J. G. Carbonell and T. M. Mitchell, Machine learning: An Artificial Intelligence Approach, Springer Science & Business Media, 2013

work page 2013
[17]

Special issue on learning from imbalanced data sets,

N. V. Chawla, N. Japkowicz and A. Kotcz, “Special issue on learning from imbalanced data sets,” ACM SIGKDD Explorations Newsletter, vol. 6, no. 1, pp. 1-6, 2004. 21

work page 2004
[18]

Evaluation: from precision, recall and F -measure to ROC, informedness, markedness and correlation,

D. M. Powers, “Evaluation: from precision, recall and F -measure to ROC, informedness, markedness and correlation,” International Journal of Machine Learning Technology, vol.2, no.1, pp.37-63, 2011

work page 2011
[19]

Wang, ed., Support Vector Machines: Theory and Applications, Springer Science & Business Media, June 2005

L. Wang, ed., Support Vector Machines: Theory and Applications, Springer Science & Business Media, June 2005

2005
[20]

Xgboost: A scalable tree boosting system ,

T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system ,” Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785-794, 2016

2016
[21]

Multivariate logistic models,

G. F. Glonek and P. McCullagh, “Multivariate logistic models,” Journal of the Royal Statistical Society: Series B (Methodological), vol.57, no.3, pp.533-546, 1995

work page 1995

[1] [1]

Covered conductor is reliable solution for electricity distribution,

M. Bruun, “Covered conductor is reliable solution for electricity distribution,” 2011, Online : www.ensto.com

2011

[2] [2]

An improved technique for online PD detection on covered conductor lines,

W. Zhang, Z. Hou, H. Li, C. Liu and N. Ma, “An improved technique for online PD detection on covered conductor lines,” IEEE Trans. on Power Delivery, vol. 29, no. 2, pp. 972-973, April 2014

2014

[3] [3]

Modeling and experimental verification of on -line PD detection in MV covered-conductor Aerial networks,

G. M. Hashmi, M. Lehtonen and M. Nordman, “Modeling and experimental verification of on -line PD detection in MV covered-conductor Aerial networks,” IEEE Trans. Dielectr. Electr. Insul., vol. 17, no.1, pp. 167–180, Feb. 2010

2010

[4] [4]

ENET Power Line Fault Detection ,

ENET Centre , “ ENET Power Line Fault Detection ,” 2018, Online: http://cenet.vsb.cz/en/informace/enet-centre/160/enet-centre-introduction.html 20

2018

[5] [5]

Available: https://www.kaggle.com/c/vsb-power-line-fault-detection

[6] [6]

Available: https://parquet.apache.org/documentation/latest/

[7] [7]

STL: A Seasonal -Trend Decomposition,

R. B. Cleveland, W. S. Cleveland, J. E. McRae and I. Terpenning, “STL: A Seasonal -Trend Decomposition,” Journal of Official Statistics, 6(1), pp.3-73, 1990

1990

[8] [8]

Sequence learning: from recognition and prediction to sequential decision making,

R. Sun and C. L. Giles, “Sequence learning: from recognition and prediction to sequential decision making,” IEEE Intelligent Systems, vol. 16, no. 4, pp. 67- 70, July-Aug. 2001

2001

[9] [9]

Forecasting the volatility of s tock price index: A hybrid model integrating LSTM with multiple GARCH -type models ,

H. Y. Kim and C. H. Won, “Forecasting the volatility of s tock price index: A hybrid model integrating LSTM with multiple GARCH -type models ,” Expert Systems with Applications , vol. 103, pp. 25 -37, Aug. 2018

2018

[10] [10]

Hourly day -ahead solar irradiance prediction u sing weather forecasts by LSTM ,

X. Qing and Y. Niu, “Hourly day -ahead solar irradiance prediction u sing weather forecasts by LSTM ,” Energy, vol. 148, pp. 461-468, Apr. 2018

2018

[11] [11]

Machine translation using deep learning: an overview,

S. P. Singh, A. Kumar, H. Darbari, L. Singh, A. Rastogi and S. Jain, “Machine translation using deep learning: an overview,” 2017 International Conference on Computer, Communications and Electronics (Comptelix), pp. 162-167, 2017

work page 2017

[12] [12]

Learning long -term dependencies with gradient descent is difficult,

Y. Bengio, P. Simard and P. Frasconi, “Learning long -term dependencies with gradient descent is difficult, ” IEEE Trans. on Neural Networks, vol. 5, pp. 157-166, 1994

1994

[13] [13]

Gradient flow in recurrent nets: the difficulty of learning long -term dependencies,

J. F. Kolen and S. C. Kremer, “Gradient flow in recurrent nets: the difficulty of learning long -term dependencies,” A Field Guide to Dynamical Recurrent Networks , IEEE, 2001

2001

[14] [14]

Long short -term memory,

S. Hochreiter and J. Schmidhuber, “Long short -term memory,” Neural computation , vol.9, pp.1735 - 1780, 1997

1997

[15] [15]

Sequence to sequence learning with neural networks,

I. Sutskever, O. Vinyals and Q. V. Le, “Sequence to sequence learning with neural networks,” Advances in Neural Information Processing Systems, pp. 3104-3112, 2014

work page 2014

[16] [16]

R. S. Michalski, J. G. Carbonell and T. M. Mitchell, Machine learning: An Artificial Intelligence Approach, Springer Science & Business Media, 2013

work page 2013

[17] [17]

Special issue on learning from imbalanced data sets,

N. V. Chawla, N. Japkowicz and A. Kotcz, “Special issue on learning from imbalanced data sets,” ACM SIGKDD Explorations Newsletter, vol. 6, no. 1, pp. 1-6, 2004. 21

work page 2004

[18] [18]

Evaluation: from precision, recall and F -measure to ROC, informedness, markedness and correlation,

D. M. Powers, “Evaluation: from precision, recall and F -measure to ROC, informedness, markedness and correlation,” International Journal of Machine Learning Technology, vol.2, no.1, pp.37-63, 2011

work page 2011

[19] [19]

Wang, ed., Support Vector Machines: Theory and Applications, Springer Science & Business Media, June 2005

L. Wang, ed., Support Vector Machines: Theory and Applications, Springer Science & Business Media, June 2005

2005

[20] [20]

Xgboost: A scalable tree boosting system ,

T. Chen and C. Guestrin, “Xgboost: A scalable tree boosting system ,” Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785-794, 2016

2016

[21] [21]

Multivariate logistic models,

G. F. Glonek and P. McCullagh, “Multivariate logistic models,” Journal of the Royal Statistical Society: Series B (Methodological), vol.57, no.3, pp.533-546, 1995

work page 1995