Pith. sign in

REVIEW 6 major objections 5 minor 33 references

Lower Dimensional Spherical Representation of Medium Voltage Load Profiles for Visualization, Outlier Detection, and Generative Modelling

T0 review · 6 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Standardized daily load profiles sit on a hidden sphere, and a principal curve through them reveals a continuous order, unifying outlier detection, clustering, and generative modeling.

desk verdict A useful practical workflow for MV load profile analysis, but the headline 'spherical structure' is a mathematical artifact of z-scoring; the real value is the arc-shaped ordering and the generative model, both of which need stronger validation. read the letter →

arxiv 2411.14346 v1 pith:LCMCIFNC submitted 2024-11-21 eess.SY cs.SY

Edgar Mauricio Salazar Duque , Bart van der Holst , Pedro P. Vergara , Juan S. Giraldo , Phuong H. Nguyen , Anne Van der Molen , Han (J.G.) Slootweg This is my paper · ORCID
classification eess.SYcs.SY
keywords loadprofilessphericalrepresentationprincipalcomponentanalysisoutlierdetectiongenerativemodellingcurvesvonMises-Fisherdistributionmediumvoltagegrids
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

This paper argues that the shapes of daily medium-voltage load profiles, once each profile is standardized by its own mean and standard deviation, live on a hypersphere. Projecting those standardized profiles with PCA into three dimensions gives a spherical cloud with an arc-shaped latent structure, and the paper shows the same pattern appearing in data from four Dutch municipalities. The authors use this single representation to replace four separate toolchains: cluster visualization, outlier detection, profile ordering, and generative modeling all happen in the same spherical coordinate system. If the claim holds, load-profile analysis becomes a problem of directional statistics rather than a patchwork of per-task methods, and continuous transitions between consumption types become directly observable and sampleable.

What carries the argument

The central object is the row-wise standardized load-profile matrix $X = \frac{1}{\sqrt{D}}\hat{P}$, whose rows have unit norm; dot products between rows therefore equal the cosines of the angles between profiles, giving the Gram matrix $S = XX^{\top}$. The argument runs through three linked pieces: PCA projects the hypersphere data into $\mathbb{R}^3$, producing an arc-shaped spherical cloud; a principal curve $f(s)$ parameterized by $s \in [0,1]$ is fitted through the cloud to recover the latent ordering; and a von Mises-Fisher distribution $f_{\mathrm{vmf}}(z; \mu_{\mathrm{vmf}}, \kappa)$ with $\mu_{\mathrm{vmf}} = f(s)$ generates new profiles that can be selected to lie between clusters. The paper's key simplifying assumption is that the three-dimensional projection is spherical rather than ellipsoidal.

What would settle it

For any of the four municipalities, fit a general ellipsoid to the first three PCA coordinates and compare residuals with the best-fit sphere from the paper's optimization; if the radius $r$ after centering shows a systematic trend with polar or azimuth angle, or its distribution is not tightly concentrated near 1, the spherical assumption and the von Mises-Fisher generative model lose support.

Watch

Extended reading notes

Core claim

The load-bearing discovery is mathematical: after row-wise standardization, each profile vector has squared norm exactly $D$, so the profiles lie on a hypersphere of radius $\sqrt{D}$; after normalization and PCA, the first three principal components yield a three-dimensional latent distribution that is approximately spherical and arc-shaped. The paper shows this structure is not an artifact of one dataset: the same eigenprofiles, radius concentration near 1, and single-mode angle distributions appear across four municipalities. On this basis it constructs a von Mises-Fisher generative model centered on a principal curve fitted to the arc, so sampling along the curve produces synthetic profiles that interpolate continuously between clusters. It also uses the fitted probability models in spherical coordinates to flag outlier profiles, including meter failures that per-meter time-series methods miss.

Load-bearing premise

The whole construction assumes the three-dimensional PCA projection is spherical rather than ellipsoidal; if the true shape is an ellipsoid, the fitted von Mises-Fisher distributions and radius-based outlier thresholds are misspecified.

Editorial extensions

If this is right

  • One spherical representation replaces separate pipelines: the same three coordinates show cluster structure, expose outliers, and serve as the sampling space for synthetic profiles.
  • Outlier detection in radius and angle coordinates catches systematic meter faults, such as meters storing absolute power values, that smoothing or per-meter historical methods cannot.
  • Each profile receives a scalar $s$ from the principal curve; clusters become intervals on $s$, and the curve gives a ranked, continuous ordering of areas by consumption similarity.
  • Sampling a von Mises-Fisher distribution centered at $f(s)$ yields profiles with continuous mixtures between clusters, such as a specified share of residential and commercial consumption, something discrete mixture models cannot do.
  • Because the spherical structure and eigenprofiles repeat across four municipalities, the directional-statistics toolkit transfers to other medium-voltage datasets.

Reading between the lines

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

  • The hypersphere property follows only from row-wise standardization, so the same latent-sphere claim should hold for any shape-focused profile dataset, including low-voltage or hourly data; checking the radius distribution on such data would test this extension directly.
  • The arc ordering is reminiscent of seriation and pseudotime problems in other fields, so the principal-curve parameter $s$ could plausibly be correlated with spatial or demographic variables, an analysis the paper does not perform.
  • The three-component reconstruction deliberately discards high-frequency structure; a practical extension would model the residual as independent noise per time step while keeping the spherical generative model for shape, improving 15-minute accuracy.
  • The 95% confidence-interval outlier rule depends on the fitted von Mises and skew-normal distributions; a nonparametric density on the sphere would provide a robustness check for the outlier labels.
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

6 major / 5 minor

Summary. The paper argues that daily medium-voltage load profiles, after row-wise z-score standardization and PCA projection into three dimensions, lie on a sphere with an arc-shaped latent distribution. On this basis it proposes a unified pipeline: fitting a sphere and spherical-coordinate marginals for outlier detection, fitting a principal curve through the arc to order profiles and define clusters, and using a von Mises-Fisher distribution centered on the principal curve for generative modeling. The claims are illustrated on load-profile data from four Dutch municipalities.

Significance. If the spherical and arc-shaped structure were established, the proposed framework would be a genuinely useful unification of visualization, outlier detection, ordering, and generation for load-profile data, with a small parameter count and stable PCA projections. The paper is also honest about several simplifications and provides a reproducible principal-curve implementation. However, the main empirical claim is weakened in two ways. First, the fact that row-wise standardized profiles lie on a hypersphere is a mathematical consequence of the normalization, not an empirical discovery; the empirical content lies entirely in the shape of the three-dimensional PCA projection. Second, the paper explicitly assumes, rather than tests, that this projection is spherical rather than ellipsoidal, even though its own Tables II and III show strong anisotropy. The proposed outlier and generative models inherit this untested geometric assumption. The work is therefore a promising demonstration, but its central latent-structure claim needs substantially more quantitative support.

major comments (6)
  1. [Section II-A, Eqs. (4)-(7)] Equation (6) as printed is incorrect: the numerator should be the sum of squared deviations, not the sum of deviations, and the denominator should be that same sum divided by D. As written, the numerator is zero by the definition of the mean. The conclusion ||p_i|| = sqrt(D) is correct after this correction, but the proof needs to be fixed. In addition, the statement that standardization 'unveils' a latent spherical distribution overstates what the algebra shows: any row-wise z-score normalized matrix lies on a hypersphere by construction. The empirical claim is specifically that the three-dimensional PCA projection has spherical and arc-shaped structure, and that claim needs to be separated from the algebraic identity.
  2. [Section III, Eq. (15)] Equation (15) is not a well-posed optimization problem as written. The objective is linear in rho with a negative coefficient, so for any center c one can make the objective arbitrarily small by taking rho arbitrarily large; no finite minimizer exists. Presumably the intended objective is a least-squares sphere fit such as sum_i ( ||z_i - c||^2 - rho )^2, where rho = r^2. Because the fitted center c and radius rho define the spherical-coordinate transformation used for all downstream outlier detection and generative modeling, this error is load-bearing and must be corrected.
  3. [Section II-A, after Eq. (14); Tables II and III] The assumption that the three-dimensional PCA projection is 'spherical-shaped instead of an ellipsoid' is never tested, and the paper's own summary statistics indicate strong anisotropy. Table II shows PC1 explaining roughly 69% of variance versus about 20% and 4% for PC2 and PC3, and Table III reports a standard deviation of about 50 degrees for the azimuthal angle versus about 9 degrees for the polar angle. These numbers are more consistent with an arc-like, effectively one-dimensional distribution than with a rotationally symmetric spherical cloud. The paper should provide a direct test of sphericity, for example by comparing the sphere-fit residuals of Eq. (15) with an ellipsoid fit, or by evaluating the eigenvalue spectrum of the local covariance structure. Until such a test is provided, the use of isotropic directional-statistics tools is not justified.
  4. [Section III, Fig. 3] The outlier detection rule fits independent von Mises marginals for phi and theta and a skew-normal marginal for r, then flags points outside the 95% confidence interval of each marginal. This ignores the strong dependence between phi and theta along the arc: points at the arc endpoints can be flagged as outliers merely because one marginal is extreme, while points that deviate from the arc in the joint (phi, theta) plane can pass all three marginal tests. A joint directional outlier criterion, for example a density threshold under a fitted VMF or Kent distribution, would be more appropriate. In addition, the 95% confidence level is chosen without justification, and no quantitative evaluation of outlier-detection performance against labeled anomalies is provided.
  5. [Section V-C, Eqs. (23)-(24)] The generative model uses a single concentration parameter kappa = 7.1, taken from the theta marginal, for a von Mises-Fisher distribution that is rotationally symmetric around its mean direction f(s). Along a curved, anisotropic arc the local spread is generally different in the tangential direction and the two normal directions, so a single kappa cannot reproduce the observed distribution. The paper should either fit an anisotropic directional distribution (such as a Kent distribution) or demonstrate empirically that the generated points reproduce the joint distribution of the projected data, rather than only the marginal summaries in Table I.
  6. [Section V-D, Tables II and III, Fig. 8] The cross-municipality validation is descriptive rather than quantitative. The paper shows that the first three eigenprofiles look similar across municipalities and that the spherical-coordinate marginals have comparable moments, but it does not test whether the three-dimensional projections in the other municipalities are spherical, nor whether the arc ordering is present. The case of Municipality 4 is particularly concerning: its three-component cumulative explained variance is only 81.8% and its outlier fraction is 16.9%, in contrast to the other municipalities. The claim that the spherical structure with latent ordering is a general property of MV load profiles requires a quantitative test, for example a goodness-of-fit measure for the spherical model or a comparison of principal-curve reconstruction error across all four datasets.
minor comments (5)
  1. [Title and Conclusion] The title contains a stray space in 'Voltage' ('V oltage'), and the Conclusion refers to the 'von Fisher-Misses distribution'; this should be 'von Mises-Fisher distribution'.
  2. [Section II-A, paragraph after Eq. (14)] There is a typo 'hypershpere' that should read 'hypersphere'.
  3. [Section IV-A, Eq. (20a)] Equation (20a) appears to omit the square in the Gaussian exponent: it should be exp(-1/2 (t - mu)^2), not exp(-1/2 (t - mu)).
  4. [Section I, paragraph after 'It is critical to notice'] The phrase 'it is possible can identify outlier profiles' is ungrammatical; it should be 'it is possible to identify'.
  5. [Section II-A, first two paragraphs] The first two paragraphs of Section II-A repeat nearly the same statement about standardization of daily MV load profiles; one of the two paragraphs should be removed or merged.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the hypersphere property is explicitly derived from the z-score normalization, the 3D spherical form is an acknowledged assumption, and the downstream ordering/outlier/generative results are empirical rather than forced by construction.

full rationale

The paper's derivation chain is largely self-contained. Eqs. (2)-(7) show algebraically that row-wise z-score standardized profiles have constant norm ||p_hat_i|| = sqrt(D); this is a direct consequence of the definition of sigma_i, not a hidden fit. The paper transparently derives it and then normalizes by sqrt(D) in Eq. (8). The potentially circular-sounding claim that the 3D PCA projection is 'spherical-shaped' is not presented as a derivation: the text explicitly says 'For simplicity, we will now assume that the projected data in the three-dimensional space have a spherical-shaped structure instead of an ellipsoid' (Section II-A). This is an admitted modeling assumption, and it is a testable limitation rather than a disguised input. The arc-shaped ordering is an empirical observation supported by the horseshoe-effect literature and by the illustrative simulation in Fig. 4; the principal-curve ordering, banded similarity matrix, and generative VMF comparisons are demonstrated on data rather than forced by the normalization identity. The only self-citations (Ref. [32] for the spherical principal-curve implementation and Ref. [33] for the MVG baseline) are to software and earlier models, and they are not used to derive the central claim by authority. Hence no step reduces, by the paper's own equations, to its inputs; the score is 0.

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

The ledger shows that the spherical representation is produced by the normalization plus several simplifying assumptions. The only data-driven ingredients are the fitted distribution parameters and the empirically observed arc; the sphere itself is not an independent discovery.

free parameters (6)
  • Sphere center c and radius rho = not reported
    Solved by the optimization in Eq. (15) on the 3D PCA projection; used to center data and define spherical coordinates.
  • Concentration kappa of the VMF generative model = 7.1
    Taken from the fitted polar angle von Mises parameters (Theta_theta), Section V-C.
  • Outlier rejection threshold = 95% confidence interval
    Chosen to define rejection regions for the von Mises and skew-normal distributions in Section III.
  • Cluster bin boundaries on s = C1 [0,0.2), C2 [0.2,0.4), C3 [0.4,0.6), C4 [0.6,1.0]
    Set post hoc to match AggHC clustering results, Section V-B.
  • Number of principal components = 3
    Selected by a 90% cumulative explained variance threshold, Section II-A.
  • Distribution parameters Theta_phi, Theta_theta, Theta_r = MLE estimates
    Used for outlier rejection regions; summary statistics are reported in Table III.
assumptions (6)
  • standard math Row-wise z-score with population standard deviation gives constant norm sqrt(D)
    Derived in Eqs. (2)-(7); this forces the hypersphere, so it is not independent empirical evidence.
  • domain assumption Projected data are distributed along orthants of the hypersphere
    Assumed in Section II-A before claiming the PCA projection forms an ellipsoid or sphere; not verified quantitatively.
  • ad hoc to paper The 3D PCA projection can be treated as a sphere rather than an ellipsoid
    Explicit 'For simplicity' assumption in Section II-A; underpins the coordinate transforms and the generative model.
  • domain assumption Arc-shaped PCoA projections arise from a banded similarity matrix and reflect a latent one-dimensional ordering
    Based on cited horseshoe-effect literature [26]-[29]; applied to load profiles as a working hypothesis.
  • domain assumption Outliers are adequately modeled by independent von Mises distributions for angles and a skew-normal for radius
    Used in Section III; independence and parametric forms are assumed without goodness-of-fit tests.
  • domain assumption Principal curve spline model and its parameters converge to a meaningful ordering
    The paper cites its own GitHub repository [32] for a robust version; no convergence or parameter details are given in the text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lower Dimensional Spherical Representation of Medium Voltage Load Profiles for Visualization, Outlier Detection, and Generative Modelling." pith.science (2026). https://pith.science/paper/LCMCIFNC

@misc{pith2026241114346,
  author       = {Pith},
  title        = {Pith review of: Lower Dimensional Spherical Representation of Medium Voltage Load Profiles for Visualization, Outlier Detection, and Generative Modelling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LCMCIFNC}},
  note         = {Machine review of arXiv:2411.14346}
}
read the original abstract

This paper presents the spherical lower dimensional representation for daily medium voltage load profiles, based on principal component analysis. The objective is to unify and simplify the tasks for (i) clustering visualisation, (ii) outlier detection and (iii) generative profile modelling under one concept. The lower dimensional projection of standardised load profiles unveils a latent distribution in a three-dimensional sphere. This spherical structure allows us to detect outliers by fitting probability distribution models in the spherical coordinate system, identifying measurements that deviate from the spherical shape. The same latent distribution exhibits an arc shape, suggesting an underlying order among load profiles. We develop a principal curve technique to uncover this order based on similarity, offering new advantages over conventional clustering techniques. This finding reveals that energy consumption in a wide region can be seen as a continuously changing process. Furthermore, we combined the principal curve with a von Mises-Fisher distribution to create a model capable of generating profiles with continuous mixtures between clusters. The presence of the spherical distribution is validated with data from four municipalities in the Netherlands. The uncovered spherical structure implies the possibility of employing new mathematical tools from directional statistics and differential geometry for load profile modelling.

Figures

Figures reproduced from arXiv: 2411.14346 by the authors.

Figure 1
Figure 1. Decomposition of subset of P in its elementary matrices (13). (a) Original subset P . (b) Standardised profiles Pˆ using (2). (c)-(e) The first three most significant elementary matrix profiles are green. i.e., X1, X2, X3, with their respective eigenvector components in a solid black line. Less significant elementary matrices, i.e., X10, X11, X12, are shown in orange, for the eigenvectors (f) v10, (g) v11, and (h) v… view at source ↗
Figure 2
Figure 2. The values of the projection Z in a 3-dimensional space. (a-c) Orthographic projection of the sphere. Each blue point represents a single transformer’s daily profile. The sphere overlayed in the data is found via (15). eigenvectors (V¯ ) results at least in an ellipsoid-shaped pro￾jection. For simplicity, we will now assume that the projected data in the three-dimensional space have a spherical-shaped structure inst… view at source ↗
Figure 3
Figure 3. Probability distributions of the spherical projection variables for the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Example of latent space ordering for the process in (20). (a1) Data [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Clustered profiles from Municipality 1 and outlier identification. (a [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Ordered dataset for Municipality 1. (a-c) The principal curve model passes through the middle of the data points representing the load profiles. It [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparison between original and synthetic profiles generated by [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Review of Smart Meter Data Analytics: Applications, Methodologies, and Challenges,

    Y . Wang, Q. Chen, T. Hong, and C. Kang, “Review of Smart Meter Data Analytics: Applications, Methodologies, and Challenges,” IEEE Trans. Smart Grid, vol. 10, no. 3, pp. 3125–3148, May 2019

    work page 2019

  2. [2]

    Guyon, Ed., Feature extraction, ser

    I. Guyon, Ed., Feature extraction, ser. Studies in fuzziness and soft computing. Berlin ; New York: Springer-Verlag, 2006, no. v. 207

    work page 2006

  3. [3]

    Model-based clustering of high- dimensional data: A review,

    C. Bouveyron and C. Brunet-Saumard, “Model-based clustering of high- dimensional data: A review,” Comput. Stat. Data Anal. , vol. 71, pp. 52–78, Mar. 2014

    work page 2014

  4. [4]

    Load profiling revisited: prosumer profiling for local energy markets,

    G. Chicco and A. Mazza, “Load profiling revisited: prosumer profiling for local energy markets,” in Local Electricity Markets. Elsevier, 2021, pp. 215–242

    work page 2021

  5. [5]

    Clustering Load Profiles for Demand Response Applications,

    S. Lin, F. Li, E. Tian, Y . Fu, and D. Li, “Clustering Load Profiles for Demand Response Applications,” IEEE Trans. Smart Grid , vol. 10, no. 2, pp. 1599–1607, Mar. 2019

    work page 2019

  6. [6]

    A review of electric load classifi- cation in smart grid environment,

    K.-l. Zhou, S.-l. Yang, and C. Shen, “A review of electric load classifi- cation in smart grid environment,” Renew. Sustain. Energy Rev., vol. 24, pp. 103–110, Aug. 2013

    work page 2013

  7. [7]

    Validating Clustering Frameworks for Electric Load Demand Profiles,

    M. Jain, T. AlSkaif, and S. Dev, “Validating Clustering Frameworks for Electric Load Demand Profiles,” IEEE Trans. Ind. Informat. , vol. 17, no. 12, pp. 8057–8065, Dec. 2021

    work page 2021

  8. [8]

    Household Energy Consumption Segmentation Using Hourly Data,

    J. Kwac, J. Flora, and R. Rajagopal, “Household Energy Consumption Segmentation Using Hourly Data,” IEEE Trans. Smart Grid , vol. 5, no. 1, pp. 420–430, Jan. 2014

    work page 2014

Show all 33 references
  1. [9]

    C-Vine Copula Mixture Model for Clustering of Residential Electrical Load Pattern Data,

    M. Sun, I. Konstantelos, and G. Strbac, “C-Vine Copula Mixture Model for Clustering of Residential Electrical Load Pattern Data,” IEEE Trans. Power Syst., vol. 32, no. 3, pp. 2382–2393, May 2017

    2017

  2. [10]

    Clustering disaggregated load profiles using a Dirichlet process mixture model,

    R. Granell, C. J. Axon, and D. C. Wallom, “Clustering disaggregated load profiles using a Dirichlet process mixture model,” Energy Convers. Manag., vol. 92, pp. 507–516, Mar. 2015

    2015

  3. [11]

    Evaluation of Dimen- sionality Reduction Techniques for Load Profiling Application in Smart Grid Environment,

    A. Aleshinloye, M. A. Manzoor, and A. Bais, “Evaluation of Dimen- sionality Reduction Techniques for Load Profiling Application in Smart Grid Environment,” IEEE Can. J. Electr. Comput. Eng. , vol. 44, no. 1, pp. 41–49, 2021

    2021

  4. [12]

    A Method to Analyzing and Clustering Aggregate Customer Load Profiles Based on PCA,

    A. Bosisio, A. Berizzi, A. Vicario, A. Morotti, B. Greco, G. Iannarelli, and D.-D. Le, “A Method to Analyzing and Clustering Aggregate Customer Load Profiles Based on PCA,” in 2020 5th Int. Conf. on Green Technology and Sust. Dev. (GTSD). Ho Chi Minh City, Vietnam: IEEE, Nov. ...

    2020

  5. [13]

    Performance analysis of dimensionality reduction techniques for demand side management,

    A. Aleshinloye, A. Bais, and I. Al-Anbagi, “Performance analysis of dimensionality reduction techniques for demand side management,” in 2017 IEEE Electrical Power and Energy Conference (EPEC) . Saska- toon, SK: IEEE, Oct. 2017, pp. 1–6

    2017

  6. [14]

    Comparisons Among Clustering Techniques for Electricity Customer Classification,

    G. Chicco, R. Napoli, and F. Piglione, “Comparisons Among Clustering Techniques for Electricity Customer Classification,” IEEE Trans. Power Syst., vol. 21, no. 2, pp. 933–940, May 2006

    2006

  7. [15]

    Com- parison of Dimensionality Reduction Techniques for Clustering and Visualization of Load Profiles,

    A. Arechiga, E. Barocio, J. J. Ayon, and H. A. Garcia-Baleon, “Com- parison of Dimensionality Reduction Techniques for Clustering and Visualization of Load Profiles,” in 2016 IEEE PES Transmission & Distribution Conference and Exposition-Latin America (PES T&D-LA) . Morelia, M...

    2016

  8. [16]

    A bottom-up framework for analysing city-scale energy data using high dimension reduction tech- niques,

    W. Khan, S. Walker, and W. Zeiler, “A bottom-up framework for analysing city-scale energy data using high dimension reduction tech- niques,” Sustainable Cities and Society , vol. 89, p. 104323, Feb. 2023

    2023

  9. [17]

    Convolutional Autoencoder Based Feature Extraction and Clustering for Customer Load Analysis,

    S. Ryu, H. Choi, H. Lee, and H. Kim, “Convolutional Autoencoder Based Feature Extraction and Clustering for Customer Load Analysis,” IEEE Trans. Power Syst. , vol. 35, no. 2, pp. 1048–1060, Mar. 2020

    2020

  10. [18]

    Automated Load Curve Data Cleansing in Power Systems,

    J. Chen, W. Li, A. Lau, J. Cao, and K. Wang, “Automated Load Curve Data Cleansing in Power Systems,” IEEE Trans. Smart Grid , vol. 1, no. 2, pp. 213–221, Sep. 2010

    2010

  11. [19]

    Detecting X- Outliers in Load Curve Data in Power Systems,

    Z. Guo, W. Li, A. Lau, T. Inga-Rojas, and K. Wang, “Detecting X- Outliers in Load Curve Data in Power Systems,” IEEE Trans. Power Syst., vol. 27, no. 2, pp. 875–884, May 2012

    2012

  12. [20]

    Electricity Theft Detection in AMI Based on Clustering and Local Outlier Factor,

    Y . Peng, Y . Yang, Y . Xu, Y . Xue, R. Song, J. Kang, and H. Zhao, “Electricity Theft Detection in AMI Based on Clustering and Local Outlier Factor,” IEEE Access, vol. 9, pp. 107 250–107 259, 2021

    2021

  13. [21]

    Real- Time Synchrophasor Data Anomaly Detection and Classification Using Isolation Forest, KMeans, and LoOP,

    E. Khaledian, S. Pandey, P. Kundu, and A. K. Srivastava, “Real- Time Synchrophasor Data Anomaly Detection and Classification Using Isolation Forest, KMeans, and LoOP,” IEEE Trans. Smart Grid, vol. 12, no. 3, pp. 2378–2388, May 2021

    2021

  14. [22]

    Using Vine Copulas to Generate Representative System States for Machine Learning,

    I. Konstantelos, M. Sun, S. H. Tindemans, S. Issad, P. Panciatici, and G. Strbac, “Using Vine Copulas to Generate Representative System States for Machine Learning,” IEEE Trans. Power Syst. , vol. 34, no. 1, pp. 225–235, Jan. 2019

    2019

  15. [23]

    Community energy storage operation via reinforcement learning with eligibility traces,

    E. M. Salazar Duque, J. S. Giraldo, P. P. Vergara, P. Nguyen, A. Van Der Molen, and H. Slootweg, “Community energy storage operation via reinforcement learning with eligibility traces,” Electric Power Systems Research, vol. 212, p. 108515, Nov. 2022

    2022

  16. [24]

    A comprehensive assessment of PV inverters operating with droop control for overvoltage mitigation in LV distribution net- works,

    P. P. Vergara, E. M. Salazar Duque, T. T. Mai, P. H. Nguyen, and H. Slootweg, “A comprehensive assessment of PV inverters operating with droop control for overvoltage mitigation in LV distribution net- works,” Renewable Energy, vol. 159, pp. 172–183, Oct. 2020

    2020

  17. [25]

    C. M. Bishop, Pattern recognition and machine learning , ser. Informa- tion science and statistics. New York: Springer, 2006

    2006

  18. [26]

    Uncovering the Horseshoe Effect in Microbial Analyses,

    J. T. Morton, L. Toran, A. Edlund, J. L. Metcalf, C. Lauber, and R. Knight, “Uncovering the Horseshoe Effect in Microbial Analyses,” mSystems, vol. 2, no. 1, pp. e00 166–16, Feb. 2017

    2017

  19. [27]

    Resemblance coefficients and the horseshoe effect in principal coordinates analysis,

    J. Podani and I. Mikl ´os, “Resemblance coefficients and the horseshoe effect in principal coordinates analysis,” Ecology, vol. 83, no. 12, pp. 3331–3343, 2002

    2002

  20. [28]

    A mathematical approach to seriation,

    D. G. Kendall, “A mathematical approach to seriation,” Philos. Trans. R. Soc. A , vol. 269, no. 1193, pp. 125–134, 1970

    1970

  21. [29]

    Uncovering pseudotemporal trajectories with covariates from single cell and bulk expression data,

    K. R. Campbell and C. Yau, “Uncovering pseudotemporal trajectories with covariates from single cell and bulk expression data,” Nat. Com- mun., vol. 9, no. 1, p. 2442, Jun. 2018

    2018

  22. [30]

    Multidimensional Scaling,

    M. A. A. Cox and T. F. Cox, “Multidimensional Scaling,” in Handbook of Data Visualization. Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp. 315–347

    2008

  23. [31]

    Principal curves,

    T. Hastie and W. Stuetzle, “Principal curves,” J. Am. Stat. Assoc., vol. 84, no. 406, pp. 502–516, 1989

    1989

  24. [32]

    E. M. S. Duque, Principal Curves for Spherical Modelling . [Online]. Available: https://github.com/MauricioSalazare/procurve

  25. [33]

    Conditional Multivariate Elliptical Copulas to Model Residential Load Profiles From Smart Meter Data,

    E. M. S. Duque, P. P. Vergara, P. H. Nguyen, A. Van Der Molen, and J. G. Slootweg, “Conditional Multivariate Elliptical Copulas to Model Residential Load Profiles From Smart Meter Data,” IEEE Trans. Smart Grid, vol. 12, no. 5, pp. 4280–4294, Sep. 2021

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.