pith. machine review for the scientific record. sign in

arxiv: 2512.15232 · v2 · submitted 2025-12-17 · 📊 stat.AP · eess.SP

Recognition: no theorem link

A Blind Source Separation Framework to Monitor Sectoral Power Demand from Grid-Scale Load Measurements

Guillaume Koechlin , Filippo Bovera , Elena Degli Innocenti , Barbara Santini , Alessandro Venturi , Simona Vazio , Piercesare Secchi
Authors on Pith no claims yet

Pith reviewed 2026-05-16 21:47 UTC · model grok-4.3

classification 📊 stat.AP eess.SP
keywords blind source separationnon-negative matrix factorizationload disaggregationsectoral power demandgrid measurementselectricity consumptiondemand flexibility
0
0 comments X

The pith

A linearly constrained non-negative matrix factorization method separates national grid load data into identifiable residential, services, and industrial components.

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

The paper introduces a blind source separation framework that disaggregates high-voltage aggregate electricity load measurements into sectoral contributions using a constrained form of non-negative matrix factorization. Prior knowledge of typical load patterns for each sector is encoded as linear constraints on the factor matrices to provide weak supervision and enable separation without direct end-use data. When applied to Italian national load records from 2021 to 2023, the method recovers distinct load profiles for residential, services, and industrial sectors and produces monthly consumption totals that align with official statistics. This capability matters because demand-side flexibility and renewable integration require visibility into how different consumer groups contribute to total load, yet collecting such data directly is expensive and incomplete at national scales. The framework is presented as applicable to other grid levels where only aggregate measurements are available.

Core claim

The central claim is that linearly-constrained non-negative matrix factorization applied to time-series aggregate load data can recover identifiable sectoral components. By embedding expected sectoral patterns as linear constraints on the basis and coefficient matrices, the factorization separates the mixed national signal into residential, services, and industrial sources. Validation on Italian 2021-2023 data shows that the resulting monthly sectoral consumption estimates remain consistent with independently reported national statistics.

What carries the argument

Linearly-constrained non-negative matrix factorization (LCNMF), which augments standard NMF by adding linear equality or inequality constraints on the factor matrices to incorporate prior sectoral load shape information and guide the blind separation.

If this is right

  • Monthly sectoral consumption can be estimated directly from open grid load measurements without additional end-use metering.
  • Sector-specific load profiles become recoverable, supporting targeted demand-response programs for different consumer groups.
  • The same constrained separation can be applied at regional or local grid scales where only aggregate data exist.
  • Results consistent with official statistics indicate the method can serve as a low-cost monitoring tool for energy planning.

Where Pith is reading between the lines

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

  • Running the separation on higher-resolution data streams could expose intra-day sectoral contributions useful for real-time balancing.
  • The linear constraints could be updated periodically from new survey data to track structural shifts in sectoral behavior.
  • Combining the disaggregated series with weather or economic covariates might improve sector-specific demand forecasting.
  • The framework could be tested on other national grids by translating local consumption statistics into the required linear constraints.

Load-bearing premise

Prior information about typical daily and weekly load patterns for each sector can be expressed as linear constraints on the factorization matrices that remain valid across the observed period and sufficient to guarantee unique separation.

What would settle it

A hold-out national dataset or independent sub-metered sectoral measurements in which the estimated monthly sectoral totals deviate systematically from reported statistics or the recovered profiles fail to match known sector-specific shapes.

Figures

Figures reproduced from arXiv: 2512.15232 by Alessandro Venturi, Barbara Santini, Elena Degli Innocenti, Filippo Bovera, Guillaume Koechlin, Piercesare Secchi, Simona Vazio.

Figure 1
Figure 1. Figure 1: Diagram showing the load decomposition process. The input data are repre [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the load disaggregation problem. In this example, each sector is [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Value of the loss L at convergence for the N = 1000 LCNMF runs. The values are vertically jittered for legibility. The grey dashed line indicates the threshold used to isolate the red cluster. As underlined in subsection 3.4, we do not choose a specific solution but we analyse and use the entire set of N∗ optimal solutions to get a robust ensemble estimation and quantify uncertainty. A first consideration … view at source ↗
Figure 4
Figure 4. Figure 4: Scree plot of the FPCA ran on 2021-2022 load curves [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sources corresponding to the N∗ = 271 retained solutions. Each thin line corresponds to one solution while the thick line, added for visualization purposes, is the geometric medoid of the thin lines, which is the closest observation to the geometric median (in terms of Euclidean distance). Since inside a sector the source number is arbitrary, the sources of household and services were aligned on the order … view at source ↗
Figure 6
Figure 6. Figure 6: Average load profiles found for the households and services sectors depending [PITH_FULL_IMAGE:figures/full_fig_p023_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of the average concentrations per day type and season across solu [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Result of the load disaggregation for the week starting on Monday January [PITH_FULL_IMAGE:figures/full_fig_p025_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of the BSS monthly sector consumption estimates with the TSO’s [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
read the original abstract

As demand-side flexibility becomes increasingly necessary to integrate variable renewable energy, understanding electricity demand composition across different grid levels is essential. However, at regional and national scales, visibility into the relative contributions of different consumer categories remains limited due to the complexity and cost of collecting end-use consumption data. To address this challenge, we propose a blind source separation framework to disaggregate open-access high-voltage grid load measurements into sectoral contributions. The approach relies on a constrained variant of non-negative matrix factorization, termed linearly-constrained non-negative matrix factorization (LCNMF), which allows prior information to be incorporated as linear constraints on the factor matrices, thereby providing weak supervision of the separation process. The framework is evaluated using Italian national load data from 2021 to 2023. Results demonstrate the identifiability of residential, services, and industrial load components and provide monthly sectoral consumption estimates consistent with reported statistics. The proposed method is generalizable and applicable to load disaggregation problems across multiple grid scales where disaggregated measurements are unavailable.

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

3 major / 2 minor

Summary. The paper proposes a linearly-constrained non-negative matrix factorization (LCNMF) framework for blind source separation of high-voltage grid load time series into residential, services, and industrial sectoral components. Prior information is encoded as linear equality constraints on the factor matrices to provide weak supervision. The method is applied to Italian national load data (2021–2023); the authors report that the three sectoral components are identifiable and that the resulting monthly consumption estimates are consistent with official statistics.

Significance. If the separation is demonstrably driven by the observed load measurements rather than by the supplied constraints, the approach would offer a practical, low-cost route to sectoral demand monitoring at grid scale without requiring end-use metering. This would be directly relevant to demand-side flexibility studies and renewable integration. The manuscript does not yet supply the quantitative diagnostics (constraint rank, null-space dimension, ablation results, or error metrics) needed to establish that the data term is load-bearing.

major comments (3)
  1. [§3.2] §3.2 (LCNMF formulation): the linear constraints are stated to encode prior sectoral patterns, yet no rank or null-space analysis of the constraint matrix is reported. Without this, it is impossible to verify that the observed load time series (rather than the constraints alone) determine the recovered factors, directly undermining the central identifiability claim.
  2. [§4] §4 (Results): the abstract and results state consistency with official statistics but supply no quantitative validation metrics (RMSE, MAPE, correlation coefficients, or confidence intervals) nor any sensitivity test that removes or perturbs subsets of the linear constraints. The reported agreement therefore cannot be assessed for statistical significance or robustness.
  3. [§3.1–3.3] §3.1–3.3: the non-negativity and linear constraints are combined in the optimization; an ablation that solves the problem with the data term removed (or with constraints only) is absent. Such a check is required to confirm that the factorization is not trivially satisfied by the constraint set.
minor comments (2)
  1. [§2] Notation for the factor matrices A and B is introduced without an explicit statement of their dimensions relative to the number of time samples and sectors; this should be clarified in §2 or §3.
  2. [Figures] Figure captions do not indicate whether error bars or uncertainty bands are shown; if none are present, this should be stated explicitly.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of identifiability and validation. We agree that additional quantitative diagnostics are needed to support the central claims and will revise the manuscript to incorporate them.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (LCNMF formulation): the linear constraints are stated to encode prior sectoral patterns, yet no rank or null-space analysis of the constraint matrix is reported. Without this, it is impossible to verify that the observed load time series (rather than the constraints alone) determine the recovered factors, directly undermining the central identifiability claim.

    Authors: We agree that a rank and null-space analysis is required to substantiate the identifiability claim. In the revised manuscript we will compute and report the rank of the constraint matrix together with the dimension of its null space, explicitly showing that the constraints alone do not determine the factors and that the data term is load-bearing. revision: yes

  2. Referee: [§4] §4 (Results): the abstract and results state consistency with official statistics but supply no quantitative validation metrics (RMSE, MAPE, correlation coefficients, or confidence intervals) nor any sensitivity test that removes or perturbs subsets of the linear constraints. The reported agreement therefore cannot be assessed for statistical significance or robustness.

    Authors: We acknowledge the absence of these metrics. The revised version will include RMSE, MAPE, and Pearson correlation coefficients between the estimated monthly sectoral loads and official statistics, together with sensitivity tests that systematically remove or perturb subsets of the linear constraints to quantify robustness. revision: yes

  3. Referee: [§3.1–3.3] §3.1–3.3: the non-negativity and linear constraints are combined in the optimization; an ablation that solves the problem with the data term removed (or with constraints only) is absent. Such a check is required to confirm that the factorization is not trivially satisfied by the constraint set.

    Authors: We will add the requested ablation study. The revised manuscript will present results obtained by solving the optimization with the data-fidelity term removed (constraints only) and compare them to the full LCNMF solution, thereby demonstrating that the observed load measurements are essential for the recovered factors. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the LCNMF derivation chain

full rationale

The paper defines LCNMF as a constrained NMF variant that incorporates external prior information as linear constraints on the factor matrices while retaining a data-fidelity term from the observed grid load time series and non-negativity. The identifiability claim and monthly estimates are presented as emerging from the joint optimization on the aggregate measurements, with consistency to reported statistics serving as an external validation rather than an input to the constraints. No self-definitional reduction, fitted-input prediction, or self-citation load-bearing step is exhibited in the abstract or described framework; the derivation remains self-contained against the load data and independent priors.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract supplies insufficient detail to enumerate specific free parameters or invented entities; the core modeling assumption is standard for NMF-based source separation.

axioms (1)
  • domain assumption Aggregate load can be expressed as a non-negative linear combination of sectoral source signals.
    Fundamental modeling choice underlying the NMF decomposition for blind source separation.

pith-pipeline@v0.9.0 · 5498 in / 1124 out tokens · 42349 ms · 2026-05-16T21:47:17.352130+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

54 extracted references · 54 canonical work pages

  1. [1]

    IEA, Key World Energy Statistics 2021, Technical Report, IEA, Paris,

    work page 2021

  2. [2]

    URL:https://www.iea.org/reports/key-world-energy-sta tistics-2021/final-consumption

    work page 2021

  3. [3]

    G. Mor, J. Cipriano, G. Martirano, F. Pignatelli, C. Lodi, F. Lazzari, B. Grillone, D. Chemisana, A data-driven method for unsupervised electricity consumption characterisation at the district level and beyond, Energy Reports 7 (2021) 5667–5684. URL:https://www.sciencedir ect.com/science/article/pii/S2352484721007988. doi:10.1016/j. egyr.2021.08.195

    work page doi:10.1016/j 2021

  4. [4]

    Voulis, M

    N. Voulis, M. Warnier, F. M. T. Brazier, Understanding spatio-temporal electricity demand at different urban scales: A data-driven approach, Applied Energy 230 (2018) 1157–1171. URL:https://www.scienced irect.com/science/article/pii/S0306261918312959. doi:10.1016/ j.apenergy.2018.08.121

    work page 2018

  5. [5]

    Langevin, J

    J. Langevin, J. L. Reyna, S. Ebrahimigharehbaghi, N. Sandberg, P. Fen- nell, C. Nägeli, J. Laverge, M. Delghust, E. Mata, M. Van Hove, J. Web- ster, F. Federico, M. Jakob, C. Camarasa, Developing a common ap- proach for classifying building stock energy models, Renewable and Sustainable Energy Reviews 133 (2020) 110276. URL:https://www. sciencedirect.com/...

    work page doi:10.1016/j.rser.2020.110276 2020

  6. [6]

    Y. Wang, Q. Chen, C. Kang, M. Zhang, K. Wang, Y. Zhao, Load profiling and its application to demand response: A review, Tsinghua Science and Technology 20 (2015) 117–129. URL:https://ieeexplore .ieee.org/document/7085625. doi:10.1109/TST.2015.7085625

    work page doi:10.1109/tst.2015.7085625 2015

  7. [7]

    Møller Andersen, P

    F. Møller Andersen, P. A. Gunkel, C. Bjerregaard, H. K. Jacobsen, Changes in hourly electricity consumption profiles and price elasticities in Denmark 2019–2022, Energy 308 (2024) 132798. URL:https://ww w.sciencedirect.com/science/article/pii/S0360544224025726. doi:10.1016/j.energy.2024.132798

    work page doi:10.1016/j.energy.2024.132798 2019

  8. [8]

    Thoma, Electrical energy usage over the business cycle, Energy Economics 26 (2004) 463–485

    M. Thoma, Electrical energy usage over the business cycle, Energy Economics 26 (2004) 463–485. URL:https://www.sciencedirect.co m/science/article/pii/S0140988304000167. doi:10.1016/j.eneco. 2004.04.006

    work page doi:10.1016/j.eneco 2004

  9. [9]

    X. Wang, C. Si, J. Gu, G. Liu, W. Liu, J. Qiu, J. Zhao, Electricity- consumption data reveals the economic impact and industry recovery during the pandemic, Scientific Reports 11 (2021) 19960. URL:https: //www.nature.com/articles/s41598-021-98259-3. doi:10.1038/s4 1598-021-98259-3, publisher: Nature Publishing Group

    work page doi:10.1038/s4 2021

  10. [10]

    Krarti, M

    M. Krarti, M. Aldubyan, Review analysis of COVID-19 impact on electricity demand for residential buildings, Renewable and Sustain- able Energy Reviews 143 (2021) 110888. URL:https://www.scie ncedirect.com/science/article/pii/S1364032121001829. doi:10.1016/j.rser.2021.110888

    work page doi:10.1016/j.rser.2021.110888 2021

  11. [11]

    van Zoest, K

    V. van Zoest, K. Lindberg, F. El Gohary, C. Bartusch, Evaluating the effects of the COVID-19 pandemic on electricity consumption patterns in the residential, public, commercial and industrial sectors in Sweden, Energy and AI 14 (2023) 100298. URL:https://www.sciencedirect. com/science/article/pii/S2666546823000708. doi:10.1016/j.egya i.2023.100298

    work page doi:10.1016/j.egya 2023

  12. [12]

    Zhang, Q

    S. Zhang, Q. Guo, R. Smyth, Y. Yao, Extreme temperatures and res- idential electricity consumption: Evidence from Chinese households, Energy Economics 107 (2022) 105890. URL:https://www.scie ncedirect.com/science/article/pii/S014098832200072X. doi:10.1016/j.eneco.2022.105890. 32

    work page doi:10.1016/j.eneco.2022.105890 2022

  13. [13]

    Gerossier, T

    A. Gerossier, T. Barbier, R. Girard, A novel method for decompos- ing electricity feeder load into elementary profiles from customer in- formation, Applied Energy 203 (2017) 752–760. URL:https://www. sciencedirect.com/science/article/pii/S0306261917308486. doi:10.1016/j.apenergy.2017.06.096

    work page doi:10.1016/j.apenergy.2017.06.096 2017

  14. [14]

    Kraft, A

    J. Kraft, A. Kraft, On the Relationship Between Energy and GNP, The Journal of Energy and Development 3 (1978) 401–403. URL:https:// www.jstor.org/stable/24806805, publisher: International Research Center for Energy and Economic Development (ICEED)

    work page arXiv 1978

  15. [15]

    F. M. Andersen, H. V. Larsen, R. B. Gaardestrup, Long term forecasting of hourly electricity consumption in local areas in Denmark, Applied Energy 110 (2013) 147–162. URL:https://www.sciencedirect.com/ science/article/pii/S0306261913003413. doi:10.1016/j.apenergy .2013.04.046

    work page doi:10.1016/j.apenergy 2013

  16. [16]

    F. M. Andersen, H. V. Larsen, T. K. Boomsma, Long-term forecasting of hourly electricity load: Identification of consumption profiles and segmentation of customers, Energy Conversion and Management 68 (2013) 244–252. URL:https://www.sciencedirect.com/science/ar ticle/pii/S0196890413000381. doi:10.1016/j.enconman.2013.01.0 18

    work page doi:10.1016/j.enconman.2013.01.0 2013

  17. [17]

    URL:http://arxiv.org/abs/2506.0

    A.Diaz-Iglesias, X.Belaunzaran, A.M.Florez-Tapia, Short-TermPower DemandForecastingforDiverseConsumerTypestoEnhanceGridPlan- ning and Synchronisation, 2025. URL:http://arxiv.org/abs/2506.0

    work page 2025

  18. [18]

    doi:10.48550/arXiv.2506.04294, arXiv:2506.04294 [cs]

    work page doi:10.48550/arxiv.2506.04294

  19. [19]

    Pagliarini, C

    G. Pagliarini, C. Bonfiglio, P. Vocale, Outdoor temperature sensitivity of electricity consumption for space heating and cooling: An application to the city of Milan, North of Italy, Energy and Buildings 204 (2019) 109512. URL:https://www.sciencedirect.com/science/article/ pii/S0378778819317360. doi:10.1016/j.enbuild.2019.109512

    work page doi:10.1016/j.enbuild.2019.109512 2019

  20. [20]

    F. M. Andersen, H. V. Larsen, N. Juul, R. B. Gaardestrup, Differenti- ated long term projections of the hourly electricity consumption in local areas. The case of Denmark West, Applied Energy 135 (2014) 523–538. URL:https://www.sciencedirect.com/science/article/pii/S030 6261914008861. doi:10.1016/j.apenergy.2014.08.075. 33

    work page doi:10.1016/j.apenergy.2014.08.075 2014

  21. [21]

    D. D. Lee, H. S. Seung, Learning the parts of objects by non-negative matrix factorization, Nature 401 (1999) 788–791. URL:https://ww w.nature.com/articles/44565. doi:10.1038/44565, number: 6755 Publisher: Nature Publishing Group

    work page doi:10.1038/44565 1999

  22. [22]

    D. D. Lee, H. S. Seung, Algorithms for non-negative matrix factoriza- tion, in: Proceedings of the 13th International Conference on Neural Information Processing Systems, NIPS’00, MIT Press, Cambridge, MA, USA, 2000, pp. 535–541

    work page 2000

  23. [23]

    J. V. Paatero, P. D. Lund, A model for generating household elec- tricity load profiles, International Journal of Energy Research 30 (2006) 273–290. URL:h t t p s : / / o n l i n e l i b r a r y . w i l e y . c o m / d o i / a b s / 1 0 . 1 0 0 2 / e r . 1 1 3 6. doi:1 0 . 1 0 0 2 / e r . 1 1 3 6, _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/er.1136

    work page doi:10.1002/er.1136 2006

  24. [24]

    Räsänen, D

    T. Räsänen, D. Voukantsis, H. Niska, K. Karatzas, M. Kolehmainen, Data-based method for creating electricity use load profiles using large amount of customer-specific hourly measured electricity use data, Ap- plied Energy 87 (2010) 3538–3545. URL:https://www.sciencedirec t.com/science/article/pii/S0306261910001832. doi:10.1016/j.ap energy.2010.05.015

    work page doi:10.1016/j.ap 2010

  25. [25]

    Teeraratkul, D

    T. Teeraratkul, D. O’Neill, S. Lall, Shape-Based Approach to Household Electric Load Curve Clustering and Prediction, IEEE Transactions on Smart Grid 9 (2018) 5196–5206. URL:https://ieeexplore.ieee. org/document/7880560. doi:10.1109/TSG.2017.2683461, conference Name: IEEE Transactions on Smart Grid

    work page doi:10.1109/tsg.2017.2683461 2018

  26. [26]

    F. M. Andersen, P. A. Gunkel, H. K. Jacobsen, L. Kitzing, Residential electricity consumption and household characteristics: An econometric analysis of Danish smart-meter data, Energy Economics 100 (2021) 105341. URL:https://www.sciencedirect.com/science/article/ pii/S0140988321002474. doi:10.1016/j.eneco.2021.105341

    work page doi:10.1016/j.eneco.2021.105341 2021

  27. [27]

    Michalakopoulos, E

    V. Michalakopoulos, E. Sarmas, I. Papias, P. Skaloumpakas, V. Mari- nakis, H. Doukas, A machine learning-based framework for clustering residential electricity load profiles to enhance demand response pro- grams, Applied Energy 361 (2024) 122943. URL:https://www.sc 34 iencedirect.com/science/article/pii/S030626192400326X. doi:10.1016/j.apenergy.2024.122943

    work page doi:10.1016/j.apenergy.2024.122943 2024

  28. [28]

    A. S. Jin, K. T. Sanders, Characterizing residential sector load curves from smart meter datasets, Applied Energy 366 (2024) 123316. URL: https://www.sciencedirect.com/science/article/pii/S0306261 924006998. doi:10.1016/j.apenergy.2024.123316

    work page doi:10.1016/j.apenergy.2024.123316 2024

  29. [29]

    Jardini, C

    J. Jardini, C. Tahan, M. Gouvea, S. Ahn, F. Figueiredo, Daily load profiles for residential, commercial and industrial low voltage consumers, IEEE Transactions on Power Delivery 15 (2000) 375–380. URL:https: //ieeexplore.ieee.org/document/847276. doi:10.1109/61.847276

    work page doi:10.1109/61.847276 2000

  30. [30]

    Bellinguer, R

    K. Bellinguer, R. Girard, A. Bocquet, A. Chevalier, ELMAS: a one-year dataset of hourly electrical load profiles from 424 French industrial and tertiary sectors, Scientific Data 10 (2023) 686. URL:https://www.na ture.com/articles/s41597-023-02542-z. doi:10.1038/s41597-023 -02542-z, publisher: Nature Publishing Group

    work page doi:10.1038/s41597-023 2023

  31. [31]

    C. Behm, L. Nolting, A. Praktiknjo, How to model European electricity load profiles using artificial neural networks, Applied Energy 277 (2020) 115564. URL:https://www.sciencedirect.com/science/article/ pii/S030626192031076X. doi:10.1016/j.apenergy.2020.115564

    work page doi:10.1016/j.apenergy.2020.115564 2020

  32. [32]

    Hart, Nonintrusive appliance load monitoring, Proceedings of the IEEE 80 (1992) 1870–1891

    G. Hart, Nonintrusive appliance load monitoring, Proceedings of the IEEE 80 (1992) 1870–1891. doi:10.1109/5.192069

    work page doi:10.1109/5.192069 1992

  33. [33]

    P. A. Schirmer, I. Mporas, Non-Intrusive Load Monitoring: A Review, IEEE Transactions on Smart Grid 14 (2023) 769–784. URL:https: //ieeexplore.ieee.org/abstract/document/9820770. doi:10.110 9/TSG.2022.3189598, conference Name: IEEE Transactions on Smart Grid

    work page arXiv 2023

  34. [34]

    Y. Zhu, S. Lu, Load profile disaggregation by Blind source separation: A wavelets-assisted independent component analysis approach, in: 2014 IEEE PES General Meeting | Conference & Exposition, 2014, pp. 1–5. URL:https://ieeexplore.ieee.org/document/6938947. doi:10.110 9/PESGM.2014.6938947, iSSN: 1932-5517

    work page arXiv 2014

  35. [35]

    Semwal, D

    S. Semwal, D. Joshi, R. S. Prasad, D. Raveendhra, The practicability of ICA in home appliances load profile separation using current signature: 35 A preliminary study, in: 2013 International Conference on Power, En- ergyandControl(ICPEC),2013, pp.756–759.URL:https://ieeexplo re.ieee.org/document/6527756. doi:10.1109/ICPEC.2013.6527756

    work page doi:10.1109/icpec.2013.6527756 2013

  36. [36]

    Herath, M

    G. Herath, M. Liyanage, C. Angammana, Assessment of Non-Intrusive Load Monitoring as a Blind Source Separation Problem, in: 2023 IEEE PES GTD International Conference and Exposition (GTD), 2023, pp. 156–160. URL:https://ieeexplore.ieee.org/document/10261743. doi:10.1109/GTD49768.2023.00056

    work page doi:10.1109/gtd49768.2023.00056 2023

  37. [37]

    Figueiredo, B

    M. Figueiredo, B. Ribeiro, A. de Almeida, Electrical Signal Source Sep- aration Via Nonnegative Tensor Factorization Using On Site Measure- ments in a Smart Home, IEEE Transactions on Instrumentation and Measurement 63 (2014) 364–373. URL:https://ieeexplore.ieee. org/document/6588559. doi:10.1109/TIM.2013.2278596, conference Name: IEEE Transactions on Ins...

    work page doi:10.1109/tim.2013.2278596 2014

  38. [38]

    Matsumoto, Y

    M. Matsumoto, Y. Fujimoto, Y. Hayashi, Energy Disaggregation Based on Semi-Binary NMF, in: P. Perner (Ed.), Machine Learning and Data Mining in Pattern Recognition, Springer International Publishing, Cham, 2016, pp. 401–414. doi:10.1007/978-3-319-41920-6_31

    work page doi:10.1007/978-3-319-41920-6_31 2016

  39. [39]

    Rahimpour, H

    A. Rahimpour, H. Qi, D. Fugate, T. Kuruganti, Non-Intrusive Energy Disaggregation Using Non-Negative Matrix Factorization With Sum-to- k Constraint, IEEE Transactions on Power Systems 32 (2017) 4430–

    work page 2017

  40. [40]

    doi:10.1109/TPWRS.2017.2660246, conference Name: IEEE Transac- tions on Power Systems

    URL:https://ieeexplore.ieee.org/document/7835299. doi:10.1109/TPWRS.2017.2660246, conference Name: IEEE Transac- tions on Power Systems

    work page doi:10.1109/tpwrs.2017.2660246 2017

  41. [41]

    Miyasawa, Y

    A. Miyasawa, Y. Fujimoto, Y. Hayashi, Energy disaggregation based on smart metering data via semi-binary nonnegative matrix factorization, Energy and Buildings 183 (2019) 547–558. URL:https://www.scienc edirect.com/science/article/pii/S0378778818302226. doi:10.101 6/j.enbuild.2018.10.030

    work page 2019

  42. [42]

    A. K. Zarabie, S. Das, An L0-Norm Constrained Non-Negative Matrix Factorization Algorithm for the Simultaneous Disaggregation of Fixed and Shiftable Loads, 2019. URL:http://arxiv.org/abs/1908.00142. doi:10.48550/arXiv.1908.00142, arXiv:1908.00142 [cs, eess]. 36

    work page doi:10.48550/arxiv.1908.00142 2019

  43. [43]

    Singh, A

    S. Singh, A. Majumdar, Analysis Co-Sparse Coding for Energy Disag- gregation, IEEE Transactions on Smart Grid 10 (2019) 462–470. URL: https://ieeexplore.ieee.org/document/8016423. doi:10.1109/TS G.2017.2743763

    work page doi:10.1109/ts 2019

  44. [44]

    D. Piga, A. Cominola, M. Giuliani, A. Castelletti, A. E. Rizzoli, Sparse Optimization for Automated Energy End Use Disaggregation, IEEE TransactionsonControlSystemsTechnology24(2016)1044–1051.URL: https://ieeexplore.ieee.org/document/7274721. doi:10.1109/TC ST.2015.2476777

    work page doi:10.1109/tc 2016

  45. [45]

    Kolter, S

    J. Kolter, S. Batra, A. Ng, Energy Disaggregation via Discriminative Sparse Coding, in: Proceedings of the 24th International Conference on Neural Information Processing Systems - Volume 1, volume 23, Curran Associates, Inc., 2010. URL:https://papers.nips.cc/paper_files /paper/2010/hash/7810ccd41bf26faaa2c4e1f20db70a71-Abstract. html. doi:10.5555/2997189.2997318

    work page doi:10.5555/2997189.2997318 2010

  46. [46]

    Wytock, J

    M. Wytock, J. Kolter, Contextually Supervised Source Separation with Application to Energy Disaggregation, Proceedings of the AAAI Con- ference on Artificial Intelligence 28 (2014). URL:https://ojs.aaai.o rg/index.php/AAAI/article/view/8769. doi:10.1609/aaai.v28i1.8 769, number: 1

    work page doi:10.1609/aaai.v28i1.8 2014

  47. [47]

    Dampeyrou, A

    C. Dampeyrou, A. Goichon, M. Ghienne, V. Tschannen, S. Schaack, Un- supervised separation of the thermosensitive contribution in the power consumption at a country scale, Applied Energy 363 (2024) 123097. URL:https://www.sciencedirect.com/science/article/pii/S030 626192400480X. doi:10.1016/j.apenergy.2024.123097

    work page doi:10.1016/j.apenergy.2024.123097 2024

  48. [48]

    Gillis, Nonnegative Matrix Factorization, Data Science, Society for Industrial and Applied Mathematics, 2020

    N. Gillis, Nonnegative Matrix Factorization, Data Science, Society for Industrial and Applied Mathematics, 2020. URL:https://epubs.siam .org/doi/book/10.1137/1.9781611976410. doi:10.1137/1.97816119 76410

    work page doi:10.1137/1.9781611976410 2020

  49. [49]

    URL:https://dati.terna.it/en/load#to tal-load

    Terna, Total Load, 2024. URL:https://dati.terna.it/en/load#to tal-load

    work page 2024

  50. [50]

    URL: https://www.terna.it/it/sistema-elettrico/statistiche/p ubblicazioni-statistiche

    Terna, Consumi 20XX, Technical Report, Terna S.p.A., 2024. URL: https://www.terna.it/it/sistema-elettrico/statistiche/p ubblicazioni-statistiche. 37

    work page 2024

  51. [51]

    Terna, Enel Household consumption Index, 2024

    work page 2024

  52. [52]

    URL:https://dati.ter na.it/en/load/imcei

    Terna, IMCEI Industrial consumption, 2024. URL:https://dati.ter na.it/en/load/imcei

    work page 2024

  53. [53]

    URL:https://dati.ter na.it/en/load/imser

    Terna, IMSER Services consumption, 2024. URL:https://dati.ter na.it/en/load/imser

    work page 2024

  54. [54]

    J. O. Ramsay, B. W. Silverman, Functional Data Analysis, Springer Series in Statistics, Springer, New York, NY, 2005. URL:http://link .springer.com/10.1007/b98888. doi:10.1007/b98888. 38

    work page doi:10.1007/b98888 2005