Pith. sign in

REVIEW 5 major objections 5 minor 83 references

Time-series surrogates from energy consumers generated by machine learning approaches for long-term forecasting scenarios

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Machine-learned generators can produce synthetic household load profiles realistic enough to replace scarce, privacy-restricted real data in long-term forecasting, and the paper identifies WGAN and MABF as the most faithful of the four…

desk verdict Useful benchmark with a real dataset, but the headline MABF/WGAN ranking is not yet secure because of an unresolved 30-minute versus 15-minute mismatch and a direct contradiction in the MMD results. read the letter →

arxiv 2506.20253 v1 pith:YGLQ7ENH submitted 2025-06-25 cs.LG cs.AI

Ben Gerhards , Nikita Popkov , Annekatrin König , Marcel Arpogaus , Bastian Schäfermeier , Leonie Riedl , Stephan Vogt , Philip Hehlert This is my paper
classification cs.LGcs.AI
keywords syntheticloadprofileslong-termforecastingWassersteinGANdenoisingdiffusionprobabilisticmodelshiddenMarkovnormalizingflowsconsumertypingenergytimeseries
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that generative machine-learning models can create synthetic household electricity load profiles realistic enough to stand in for real data in long-term forecasting and grid-planning applications, while preserving privacy. On an open dataset of roughly 550 German households at 15-minute resolution, it compares four generators—a Wasserstein GAN, a denoising diffusion model, a hidden Markov model, and a masked autoregressive Bernstein polynomial normalizing flow—against the standard load profiles utilities typically use. The paper reports that the WGAN and MABF perform best across error, distributional, and structural similarity metrics, and that every ML generator outperforms the standard load profile baseline. The practical point is to give practitioners a principled choice of generator for long-horizon synthetic load data, since real smart-meter data are scarce and legally restricted.

What carries the argument

The argument is carried by a four-model generative toolkit and a multi-metric evaluation battery. The WGAN uses an LSTM generator and a feed-forward critic, both conditioned on categorical and continuous metadata such as time embeddings, temperature, and consumer-cluster identities, and is trained with the Wasserstein distance and gradient penalty. The DDPM learns to reverse a noise-adding process with a 1D U-Net as noise predictor. The HMM self-classifies consumption into hidden states and samples transitions per weekday and season. The MABF factorizes the joint density autoregressively and uses monotone Bernstein polynomial transformations to map strictly positive load values to a Gaussian base distribution, with masking enforcing the autoregressive property. The evaluation battery—visual diagnostics, seasonal typical weeks, UMAP projections of daily samples, consumer-type clustering, correlation and similarity indices, and the error metrics MAE, MAPE, RMSE, MMD, Pearson correlation, and SSIM—is what lets the paper rank the generators, with WGAN and MABF winning across the battery rather than on any single metric.

What would settle it

Open the training configuration for the MABF and check whether the modeled dimension $D$ is 48, which the methods section ties to 30-minute intervals, rather than 96, the native 15-minute daily length. If $D=48$ was used, retrain MABF at $D=96$ on the same data and recompute the Table 2 metrics; if MABF no longer ties WGAN on MAE, RMSE, MMD, Pearson correlation, and SSIM, the paper's headline ranking is falsified.

Watch

Extended reading notes

Core claim

On a country-wide German household dataset with 15-minute resolution and at least 2.5 years of records per consumer, the paper trains four synthetic time-series generators and compares them over a one-year horizon against the original profiles. Its central finding is that the conditional Wasserstein GAN and the masked autoregressive Bernstein normalizing flow produce the most faithful surrogates: they preserve median consumption, daily and weekly rhythms, seasonal variation, and consumer-type structure, and they achieve the lowest MAE, MAPE, RMSE, and MMD alongside the highest Pearson correlation and SSIM. The DDPM captures seasonality well but overestimates consumption and gravitates toward the most common consumer types; the HMM follows the average closely but compresses variation; and the standard load profile baseline is the least representative, occupying clusters that real households almost never occupy. All ML models are judged to outperform the standard load profile on every tested error and similarity metric.

Load-bearing premise

The ranking depends on all four models being trained and evaluated on the same 15-minute data with comparable effort; the MABF description says $D = 48$ for 30-minute intervals, so if that was literally true, the comparison is not on equal footing.

Editorial extensions

If this is right

  • Grid planners and state estimators can generate long-horizon synthetic residential load profiles that preserve daily, weekly, seasonal, and consumer-type structure instead of relying on standard load profiles.
  • The DDPM is a reasonable choice when seasonal fidelity matters most, but its consumption overestimation and bias toward common consumer types must be accounted for in downstream applications.
  • The HMM offers interpretability and simplicity but produces low-variation, type-biased surrogates, so it is better suited to analytical quality assessment than to realistic data generation.
  • Because the surrogates show low pairwise correlation with their original consumers while preserving type-level statistics, they can serve as anonymized, privacy-preserving stand-ins for real load data.
  • The standard load profile baseline is not representative of the studied households and would mislead long-term individual-consumer planning, reinforcing the need for learned generators.

Reading between the lines

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

  • Inference: The reported ranking could shift if the MABF was actually run at 30-minute resolution (with $D=48$) while the other models used the native 15-minute data; retraining MABF at $D=96$ on the same data is a direct, cheap test of the headline result.
  • Inference: Because the dataset covers around 550 households over up to 2.5 years, the ranking is a benchmark for current German private households and may not transfer to other countries, commercial or industrial consumers, or future load mixtures with heat pumps, EVs, and battery storage.
  • Inference: The finding that surrogates are type-accurate but not individual-accurate suggests downstream tasks needing per-household behavior, such as local grid constraints, may require different validation than aggregate planning tasks.
  • Inference: A natural extension is a standardized synthetic-load benchmark that fixes this evaluation battery and adds membership-inference or nearest-neighbor re-identification tests to certify the privacy claim rather than assuming it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper presents a comparative evaluation of four generative models—a hybrid Wasserstein GAN (WGAN), a Denoising Diffusion Probabilistic Model (DDPM), a Hidden Markov Model (HMM), and a Masked Autoregressive Bernstein-polynomial normalizing Flow (MABF)—for synthesizing household electricity load profiles from the OpenMeter dataset (15-minute resolution, 552 households). The authors also compare against standard load profiles (SLP). They assess quality through visual diagnostics, seasonality, UMAP overlap, consumer-type clustering, correlation analysis, and error/similarity metrics, and conclude that WGAN and MABF perform best, that all ML models outperform SLP, and that the generated surrogates are suitable for long-term forecasting scenarios. The paper includes a data availability statement and makes the curated/generated data publicly available.

Significance. If the central ranking is reliable, the paper would be a useful empirical benchmark for generative load-profile modeling, with clear practical implications for grid planning and privacy-preserving data sharing. Its strengths are the use of a real open dataset, the breadth of compared model families (GAN, diffusion, HMM, normalizing flow), and the multi-criteria evaluation protocol. The data release is a concrete reproducibility asset. However, the main conclusion—that WGAN and MABF are the best models—rests on several unresolved internal inconsistencies and on an evaluation protocol that appears circular for the consumer-typing analysis. The manuscript therefore does not yet support its headline ranking in its current form.

major comments (5)
  1. [§8.6.4 vs §8.4 and §8.6.5] The MABF is defined with "D = 48 for all 30-minute intervals of a single day," whereas the dataloader in §8.4 uses L = 96 at 15-minute resolution and the HMM section (§8.6.5) explicitly uses 96 values per day. If MABF was trained and evaluated on 48-dimensional half-hourly daily samples, then its MAE, RMSE, MMD, and similarity scores in Table 2 and Figure S7 are not directly comparable with those of WGAN, DDPM, and HMM. Since the claim that MABF belongs at the top of the ranking depends on these metrics, the authors must either document that MABF was evaluated at 96 points per day at 15-minute resolution, or rerun the comparison at a common resolution.
  2. [§4.6 vs Figure S7] Section 4.6 states that WGAN and MABF achieved the lowest MMD and best similarity metrics, citing Figure S7, but the Figure S7 caption says "DDPM and WGAN achieved the highest average similarity. MABF showed a markedly reduced similarity, whereas the SLP samples had the lowest resemblance to the real distribution." These statements are mutually contradictory with respect to MABF. The authors need to reconcile them and report the exact MMD estimator, kernel, bandwidth, and sample sizes used; without this, the MMD-based part of the ranking is not verifiable.
  3. [§4.1 vs Table 1] The text says "The MABF exhibited a slight shift that resulted in +10% median power consumption." Table 1 reports median values of 0.35 for OM and 0.44 for MABF, which is a shift of about +25.7%, whereas the means 0.57 and 0.63 correspond to about +10.5%. The statement in the text should be corrected to refer to the mean if that is the intended statistic, or the table values should be reconciled.
  4. [§8.3, §8.4, and §4.4] The consumer-typing evaluation is circular. The dataloader in §8.4 feeds cluster-id and sensor-id as conditioning inputs, and the model is then scored in §4.4 on how often generated profiles fall into the same consumer-type cluster as their paired original consumer. A model that simply follows its conditioning inputs will trivially score highly. The authors should either evaluate unconditional or out-of-condition generation, or clearly separate the conditioning inputs from the clustering labels used for evaluation. In addition, the paired-generation protocol for the 552 consumers is not specified: how many samples per consumer, what conditions are fixed, and how the pairing is defined.
  5. [Abstract, §1, and §6] The paper is framed around "long-term forecasting scenarios," but no forecasting experiment is conducted and no downstream task is evaluated. The contribution is the generation and fidelity assessment of synthetic profiles; the claimed suitability for long-term forecasting is an assertion, not a demonstrated result. A minimal forecasting evaluation—for example, training a simple forecaster on synthetic data and testing on real data—would substantially strengthen the claim, or the framing should be narrowed to synthetic profile generation.
minor comments (5)
  1. [§8.6.5 and §8.6.6] Sections 8.6.5 and 8.6.6 are identical duplicate subsections on Hidden Markov models; one should be removed.
  2. [§2.3] The metric called "mean absolute squared error (MASE)" is misnamed; MASE stands for mean absolute scaled error. The intended metric should be named correctly or replaced by the appropriate formula.
  3. [§8.6.5 heading] The heading "Hidden Markov models (HHMs)" contains a typo; it should read "HMMs."
  4. [§8.2.1] The MMD formula in §8.2.1 is given without specifying the kernel k or its bandwidth; this is necessary for reproducibility and for interpreting the MMD values in Figure S7.
  5. [§4.4] The claim that clustering "revealed fifteen stable consumer types" is not supported by any stability analysis (e.g., bootstrap or silhouette measures); the notion of stability should be defined or replaced with a descriptive statement.

Circularity Check

1 steps flagged · score 4.0 of 10

Partial circularity in the consumer-typing axis: models are conditioned on cluster-id and sensor-id (§8.4, §8.6.2) and then scored on how often their surrogates fall into those same clusters (§4.4), so that axis is statistically forced; the headline error-metric ranking is independent but is weakened by an MMD claim that contradicts the Figure S7 caption and by MABF's ambiguous D=48 resolution.

  1. fitted input called prediction [§4.4 (Consumer typing and stability); §8.4 (Dataloader); §8.6.2 (Conditional WGAN)]
    ""Each training sample is represented as a three-tuple: 1. categorical conditions ( cluster-id and sensor-id ) ... 3. power consumption time series" (Dataloader) ... "Additionally the cluster-id that the sensor falls into is also used." (Conditional WGAN) ... "we used the centroids from the consumer typing of the OM data and assigned the surrogate power profiles from our various ML-models to the respective consumer types (Figure S5). The digital surrogates from the WGAN and MABF exhibited a distribution of consumer types that was very close to the OM data (Figure 6a)." (Consumer typing)"

    The consumer-type preservation claim is measured by cluster membership, and the cluster label is an input. The same OM-data clustering (PCA+k-means, §8.1, [19]) that defines the consumer types is used (i) to build the cluster-id (and sensor-id) conditioning fed to the generative models and (ii) as the evaluation scaffold: each surrogate is assigned to the nearest OM centroid, then scored on how close the resulting cluster distribution is to OM and how often the surrogate joins its paired consumer's cluster.

full rationale

The consumer-typing axis of Section 4.4 (Figures 6a/6b, 8b, S5) is genuinely circular: the dataloader feeds cluster-id and sensor-id as conditioning inputs (§8.4), the conditional WGAN confirmedly uses them at generation (§8.6.2, 'the cluster-id that the sensor falls into is also used'), and the evaluation then assigns each surrogate to the same OM clustering and reports that WGAN/MABF outputs mirror the OM consumer-type distribution and match their paired consumer's cluster. For any model that respects its conditioning, the output cluster equals the conditioned cluster, so these results largely verify conditioning adherence rather than a learned representation of consumer types; the input label distribution and the evaluated outcome distribution are the same object. This bias is partial because the paper's headline ranking in Section 4.6 rests on paired error and similarity metrics (MAE, MAPE, RMSE, Pearson, SSIM, MMD) computed between surrogates and their real counterparts, which are external comparisons and not circular. Two non-circular correctness risks nonetheless undercut that headline and should be weighed in the verdict: (i) Section 8.6.4 defines MABF with D=48 'for all 30-minute intervals of a single day' while the OpenMeter data, the dataloader (L=96 at 15min), and the HMM section all use 96 values per day, so MABF's evaluation resolution is ambiguous and its top ranking in Table 2 is not clearly equal-footing with the other models; (ii) Section 4.6 states WGAN and MABF achieved the lowest MMD, but the Figure S7 caption says 'DDPM and WGAN achieved the highest average similarity. MABF showed a markedly reduced similarity,' which cannot both hold without an unstated caveat. Self-citations ([19] for clustering, [75] for Bernstein-polynomial normalizing flows, [58] for the data release) describe methods that are also specified in the text and are not invoked as external uniqueness theorems, so they are not load-bearing circularity. Overall: one supporting evaluation axis reduces to its conditioning inputs, while the core model comparison retains independent content; the central claim is therefore partially but not wholly circular, and its security is further weakened by the internal inconsistencies noted above.

Assumptions & free parameters 10 free parameters · 6 assumptions · 0 invented entities

The central comparison relies on several hand-chosen settings (clustering k=15, PCA=5, HMM states=50, WGAN defaults, UMAP parameters, data filters) and on domain assumptions about the OpenMeter labels, the conditioning inputs, and the privacy interpretation. MABF and DDPM hyperparameters are partly unreported. No invented entities are introduced.

free parameters (10)
  • Number of consumer clusters k = 15
    Chosen by hand for k-means++ on PCA components; drives the consumer-typing evaluation and the cluster-id conditioning input.
  • Number of PCA components = 5
    PC1-5 used for clustering typical weeks, following reference [19]; affects the ground-truth consumer labels.
  • HMM state count = 50
    Each weekday-season HMM is initialized with 50 hidden states; the choice is stated without a search.
  • WGAN gradient penalty coefficient lambda = 10
    Default from Gulrajani et al. [45], used for all WGAN training; a standard but manually chosen hyperparameter.
  • WGAN critic iterations per generator update = 5
    Default in Algorithm 1; affects training stability and final fidelity.
  • WGAN learning rate = 0.001
    Adam learning rate in Algorithm 1; no tuning reported.
  • UMAP hyperparameters = n_neighbors=10, min_dist=1
    Chosen from pilot experiments to optimize separation and overlap of real and synthetic samples in Figure 4.
  • Data filtering thresholds = minimum 2.5 years, maximum 5% NaN
    Inclusion criteria for OpenMeter profiles in Section 2.1; the exact thresholds are choices that shape the dataset.
  • MABF Bernstein polynomial order M
    Not reported in the manuscript, despite controlling the capacity of the MABF transformation.
  • DDPM training hyperparameters
    Diffusion steps, learning rate, batch size, and noise schedule are not reported, leaving the DDPM comparison under-specified.
assumptions (6)
  • standard math Change-of-variable formula for normalizing flows (Eq. 5)
    Used to compute densities and train MABF; standard, unproved background.
  • standard math Wasserstein distance with gradient penalty provides stable GAN training
    Follows Arjovsky et al. [44] and Gulrajani et al. [45]; standard in the literature.
  • domain assumption Consumer typing via PCA on typical weeks and k-means (k=15) is a valid ground truth
    The evaluation of consumer-type preservation depends on these labels being meaningful; the clustering is taken from reference [19] without validation on this dataset.
  • domain assumption Cluster-id and sensor-id can be used as conditioning inputs without leaking the evaluation target
    Section 8.4 feeds these labels into WGAN, DDPM, and MABF; the typing evaluation assumes this does not trivially determine the cluster match.
  • domain assumption Temperature from ERA5 is a sufficient exogenous driver for seasonal consumption patterns
    Used as a condition in all models; no sensitivity analysis is given.
  • domain assumption Low correlation between real and synthetic profiles implies anonymization
    Sections 4.5 and 6 infer GDPR-compliant anonymization from correlation statistics without a re-identification attack.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Time-series surrogates from energy consumers generated by machine learning approaches for long-term forecasting scenarios." pith.science (2026). https://pith.science/paper/YGLQ7ENH

@misc{pith2026250620253,
  author       = {Pith},
  title        = {Pith review of: Time-series surrogates from energy consumers generated by machine learning approaches for long-term forecasting scenarios},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YGLQ7ENH}},
  note         = {Machine review of arXiv:2506.20253}
}
read the original abstract

Forecasting attracts a lot of research attention in the electricity value chain. However, most studies concentrate on short-term forecasting of generation or consumption with a focus on systems and less on individual consumers. Even more neglected is the topic of long-term forecasting of individual power consumption. Here, we provide an in-depth comparative evaluation of data-driven methods for generating synthetic time series data tailored to energy consumption long-term forecasting. High-fidelity synthetic data is crucial for a wide range of applications, including state estimations in energy systems or power grid planning. In this study, we assess and compare the performance of multiple state-of-the-art but less common techniques: a hybrid Wasserstein Generative Adversarial Network (WGAN), Denoising Diffusion Probabilistic Model (DDPM), Hidden Markov Model (HMM), and Masked Autoregressive Bernstein polynomial normalizing Flows (MABF). We analyze the ability of each method to replicate the temporal dynamics, long-range dependencies, and probabilistic transitions characteristic of individual energy consumption profiles. Our comparative evaluation highlights the strengths and limitations of: WGAN, DDPM, HMM and MABF aiding in selecting the most suitable approach for state estimations and other energy-related tasks. Our generation and analysis framework aims to enhance the accuracy and reliability of synthetic power consumption data while generating data that fulfills criteria like anonymisation - preserving privacy concerns mitigating risks of specific profiling of single customers. This study utilizes an open-source dataset from households in Germany with 15min time resolution. The generated synthetic power profiles can readily be used in applications like state estimations or consumption forecasting.

Figures

Figures reproduced from arXiv: 2506.20253 by the authors.

Figure 1
Figure 1. DDPM and HMM principles - a) Gradually adding noise to a sample as an [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Distribution of power consumption - Absolute frequency of power consumption [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Daily power consumption - Shown are daily averages [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Distribution of daily samples - UMAP analysis of single day samples. 2D [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Classification of OM consumers - Shown are the average typical weeks ( [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Consumer-type distribution - Note: y-axis is log-scaled. a) Initial clustering with [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Mixed consumer typing - Shown are the average typical weeks (+ 95% CI) for [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Consumer-type distributions - Clustering with all power profiles (all models, [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Similarity analysis of power profiles - Correlation analysis of power consumption [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: Schematic for an embedding - Example of how an embedding would encode [PITH_FULL_IMAGE:figures/full_fig_p023_10.png]
Figure 11
Figure 11. Figure 11: Encoding of time embeddings - a) Data points as sine and cosine over a 24h [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]
Figure 12
Figure 12. Figure 12: Scheme of the Condition Block 8.5. Synthesis of power profiles This section provides more detailed and in depth information about the models used. 8.5.1. Hybrid Wasserstein GAN (hybridWGAN) The WGAN used in this paper is implemented using an LSTM as the generator and …
Figure 13
Figure 13. Figure 13: Scheme of the WGAN with conditions 8.6.2. Conditional WGAN Another approach is to use the aforementioned meta-data that is avail￾able in the dataset. We selected the meta-data categories that are probably most useful for the model to improve realism of the generated p…
Figure 14
Figure 14. Figure 14: Schematic of the DDPM framework. Inputs are samples, a variety of condi [PITH_FULL_IMAGE:figures/full_fig_p032_14.png]
Figure 15
Figure 15. Figure 15: Schematic of the DDPM UNET-architecture and building blocks. [PITH_FULL_IMAGE:figures/full_fig_p033_15.png]
Figure 16
Figure 16. Figure 16: Schematic of the proposed MABF model for [PITH_FULL_IMAGE:figures/full_fig_p036_16.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

83 extracted references · 51 canonical work pages

  1. [2]

    F. v. Scheidt, H. Medinov´ a, N. Ludwig, B. Richter, P. Staudt, C. Wein- hardt, Data analytics in the electricity sector – a quantitative and qualitative literature review, Energy and AI 1 (2020-08-01) 100009. doi:10.1016/j.egyai.2020.100009. URL https://www.sciencedirect.com/science/article/pii/S2666546820300094

    arXiv 2020

  2. [3]

    Kim, S.-B

    T.-Y. Kim, S.-B. Cho, Predicting residential energy consumption using CNN-LSTM neural networks, Energy 182 (2019-09) 72–81. doi:10.1016/j.energy.2019.05.230. URL https://linkinghub.elsevier.com/retrieve/pii/S0360544219311223

    work page doi:10.1016/j.energy.2019.05.230 2019

  3. [4]

    J. Q. Wang, Y. Du, J. Wang, LSTM based long-term energy con- sumption prediction with periodicity, Energy 197 (2020-04) 117197. doi:10.1016/j.energy.2020.117197. URL https://linkinghub.elsevier.com/retrieve/pii/S0360544220303042

    work page arXiv 2020

  4. [5]

    Salinas, V

    D. Salinas, V. Flunkert, J. Gasthaus, T. Januschowski, DeepAR: Probabilistic forecasting with autoregressive recurrent networks, International Journal of Forecasting 36 (3) (2020-07) 1181–1191. doi:10.1016/j.ijforecast.2019.07.001. URL https://linkinghub.elsevier.com/retrieve/pii/S0169207019301888

    doi:10.1016/j.ijforecast.2019.07.001 2020

  5. [6]

    J. M. Santos-Herrero, J. M. Lopez-Guede, I. Flores Abascal, E. Zulueta, Energy and thermal modelling of an office building to develop an artifi- cial neural networks model, Sci Rep 12 (1) (2022-05-27) 8935, publisher: 39 Nature Publishing Group. doi:10.1038/s41598-022-12924-9. URL https://www.nature.com/articles/s41598-022-12924-9

    work page doi:10.1038/s41598-022-12924-9 2022

  6. [7]

    Aryandoust, A

    A. Aryandoust, A. Patt, S. Pfenninger, Enhanced spatio-temporal elec- tric load forecasts using less data with active deep learning, Nat Mach Intell 4 (11) (2022-11) 977–991, publisher: Nature Publishing Group. doi:10.1038/s42256-022-00552-x. URL https://www.nature.com/articles/s42256-022-00552-x

    work page doi:10.1038/s42256-022-00552-x 2022

  7. [8]

    Monitoringbericht 2023 monitoringbericht gem¨ aߧ 63 abs. 3 i. v. m. § 35 EnWG und § 48 abs. 3 i. v. m. § 53 abs. 3 GWB (2023-11-29)

    2023

  8. [9]

    URL http://data.europa.eu/eli/reg/2016/679/oj

    Regulation (EU) 2016/679 of the european parliament and of the council of 27 april 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing directive 95/46/EC (general data protection regulation) (text with EEA relevance) (2016-05-04). URL http://data.europa.eu/eli/reg/20...

    2016

Show all 83 references
  1. [10]

    Kezunovic, L

    M. Kezunovic, L. Xie, S. Grijalva, The role of big data in improv- ing power system operation and protection, in: 2013 IREP Sympo- sium Bulk Power System Dynamics and Control - IX Optimization, Security and Control of the Emerging Power Grid, 2013-08, pp. 1–9. doi:10.1109/IREP...

    2013

  2. [11]

    N. Yu, S. Shah, R. Johnson, R. Sherick, M. Hong, K. Loparo, Big data analytics in power distribution systems, in: 2015 IEEE Power & Energy Society Innovative Smart Grid Technologies Conference (ISGT), 2015- 02, pp. 1–5. doi:10.1109/ISGT.2015.7131868. URL https://ieeexplore.iee...

    2015

  3. [12]

    Albrecht, openMeter data platform (2024)

    M. Albrecht, openMeter data platform (2024). URL https://www.openmeter.de/

    2024

  4. [13]

    Meier, C

    H. Meier, C. F¨ unfgeld, T. Adam, B. Schieferdecker, Repr¨ asentative VDEW-lastprofile [representative VDEW load profiles] (1999). URL https://www.bdew.de/energie/standardlastprofile-strom/ 40

    1999

  5. [14]

    I. J. Ben ´ ıtez S´ anchez, A. Quijano Lopez, I. Delgado Espinos, J. L. Diez Ruano, Classification of customers based on temporal load pro- file patterns, Cigre Science & engineering 7 (2017) 143–148

    2017

  6. [15]

    R¨ as¨ anen, M

    T. R¨ as¨ anen, M. Kolehmainen, Feature-based clustering for electricity use time series data, in: Adaptive and Natural Computing Algorithms: 9th International Conference, ICANNGA 2009, Kuopio, Finland, April 23-25, 2009, Revised Selected Papers 9, Springer, 2009, pp. 401–412

    2009

  7. [16]

    Yilmaz, J

    S. Yilmaz, J. Chambers, M. K. Patel, Comparison of cluster- ing approaches for domestic electricity load profile characterisation- implications for demand side management, Energy 180 (2019) 665–677

    2019

  8. [17]

    Y. Shi, T. Yu, Q. Liu, H. Zhu, F. Li, Y. Wu, An approach of electrical load profile analysis based on time series data mining, IEEE Access 8 (2020) 209915–209925

    2020

  9. [18]

    Silipo, P

    R. Silipo, P. Winters, Big data, smart energy, and predictive analytics, Time Series Prediction of Smart Energy Data 1 (37) (2013)

    2013

  10. [19]

    Riedl, M

    L. Riedl, M. Braun, P. Hehlert, Efficient identification of customer types in energy consumption data: Leveraging dimensionality reduction and k-means clustering methods, in: CIGRE, Paris Session 2024, Ref C6- 11443-2024, 2024, pp. 1–19

    2024

  11. [20]

    T. Hong, P. Pinson, Y. Wang, R. Weron, D. Yang, H. Zareipour, Energy forecasting: A review and outlook, IEEE Open J. Power Energy 7 (2020) 376–388. doi:10.1109/OAJPE.2020.3029979. URL https://ieeexplore.ieee.org/document/9218967/

    2020

  12. [21]

    Gretton, K

    A. Gretton, K. Borgwardt, M. Rasch, B. Sch¨ olkopf, A. Smola, A kernel method for the two-sample-problem, Advances in neural information processing systems 19 (2006)

    2006

  13. [22]

    McInnes, J

    L. McInnes, J. Healy, J. Melville, Umap: Uniform manifold approxima- tion and projection for dimension reduction. arxiv 2018, arXiv preprint arXiv:1802.03426 10 (1802)

    2018 arXiv

  14. [23]

    I. K. Nti, M. Teimeh, O. Nyarko-Boateng, A. F. Adekoya, Electricity load forecasting: a systematic review, Journal of Electrical Systems and 41 Inf Technol 7 (1) (2020-09-09) 13. doi:10.1186/s43067-020-00021-8. URL https://doi.org/10.1186/s43067-020-00021-8

    2020 doi

  15. [24]

    G. E. P. Box, G. M. Jenkins, G. C. Reinsel, G. M. Ljung, Time series analysis: forecasting and control, fifth edition Edition, Wiley series in probability and statistics, John Wiley & Sons, Inc

  16. [25]

    Li, Energy consumption forecasting with deep learning, J

    Y. Li, Energy consumption forecasting with deep learning, J. Phys.: Conf. Ser. 2711 (1) (2024-02-01) 012012. doi:10.1088/1742- 6596/2711/1/012012. URL https://iopscience.iop.org/article/10.1088/1742-6596/2711/1/012012

    2024 doi

  17. [26]

    E. M. De Oliveira, F. L. Cyrino Oliveira, Forecasting mid-long term electric energy consumption through bagging ARIMA and exponential smoothing methods, Energy 144 (2018-02) 776–788. doi:10.1016/j.energy.2017.12.049. URL https://linkinghub.elsevier.com/retrieve/pii/S0360544217320820

    2018 doi

  18. [27]

    B. Dong, C. Cao, S. E. Lee, Applying support vector machines to predict building energy consumption in tropical region, Energy and Buildings 37 (5) (2005-05) 545–553. doi:10.1016/j.enbuild.2004.09.009. URL https://linkinghub.elsevier.com/retrieve/pii/S0378778804002981

    2005 doi

  19. [28]

    J. Jiao, H. Brugger, M. Behrisch, W. Eichhammer, Identifying drivers of residential energy consumption by explainable energy demand fore- casting, ECEEE SUMMER STUDY PROCEEDINGS (2022)

    2022

  20. [29]

    G. K. Tso, K. K. Yau, Predicting electricity energy consumption: A comparison of regression analysis, decision tree and neural networks, Energy 32 (9) (2007-09) 1761–1768. doi:10.1016/j.energy.2006.11.010. URL https://linkinghub.elsevier.com/retrieve/pii/S0360544206003288

    2007 doi

  21. [30]

    C. Li, Z. Ding, D. Zhao, J. Yi, G. Zhang, Building energy consumption prediction: An extreme deep learning approach, Energies 10 (10) (2017- 10-07) 1525. doi:10.3390/en10101525. URL https://www.mdpi.com/1996-1073/10/10/1525

    2017 doi

  22. [31]

    F. G. Y. Souhe, C. F. Mbey, A. T. Boum, P. Ele, FORECAST- ING OF ELECTRICAL ENERGY CONSUMPTION OF HOUSE- HOLDS IN a SMART GRID, IJEEP 11 (6) (2021-11-05) 221–233. 42 doi:10.32479/ijeep.11761. URL https://econjournals.com/index.php/ijeep/article/view/11761

    2021 doi

  23. [32]

    Singh, A

    S. Singh, A. Yassine, Big data mining of energy time series for behavioral analytics and energy consumption forecasting, Energies 11 (2) (2018-02-

    2018

  24. [33]

    doi:10.3390/en11020452

    452. doi:10.3390/en11020452. URL https://www.mdpi.com/1996-1073/11/2/452

    1996 doi

  25. [34]

    Muralitharan, R

    K. Muralitharan, R. Sakthivel, R. Vishnuvarthan, Neural network based optimization approach for energy demand prediction in smart grid, Neu- rocomputing 273 (2018-01) 199–208. doi:10.1016/j.neucom.2017.08.017. URL https://linkinghub.elsevier.com/retrieve/pii/S0925231217313681

    2018 doi

  26. [35]

    Rahman, V

    A. Rahman, V. Srikumar, A. D. Smith, Predicting electricity consumption for commercial and residential buildings using deep recurrent neural networks, Applied Energy 212 (2018-02) 372–385. doi:10.1016/j.apenergy.2017.12.051. URL https://linkinghub.elsevier.com/retrieve/pii/S03...

    2018 doi

  27. [36]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need (2017-12- 05). arXiv:1706.03762 [cs]. URL http://arxiv.org/abs/1706.03762

    2017 arXiv

  28. [37]

    S. Li, X. Jin, Y. Xuan, X. Zhou, W. Chen, Y.-X. Wang, X. Yan, En- hancing the locality and breaking the memory bottleneck of transformer on time series forecasting, 33rd Conference on Neural Information Pro- cessing Systems NeurIPS 2019 (2019)

    2019

  29. [38]

    Qureshi, M

    M. Qureshi, M. A. Arbab, S. u. Rehman, Deep learning-based fore- casting of electricity consumption, Sci Rep 14 (1) (2024-03-18) 6489, publisher: Nature Publishing Group. doi:10.1038/s41598-024-56602-4. URL https://www.nature.com/articles/s41598-024-56602-4

    2024 doi

  30. [39]

    Lai, W.-C

    G. Lai, W.-C. Chang, Y. Yang, H. Liu, Modeling long- and short-term temporal patterns with deep neural networks, in: The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, SIGIR ’18, Association for Computing Machinery, 2018-06- 27, pp....

    2018

  31. [40]

    P. V. B. Ramos, S. M. Villela, W. N. Silva, B. H. Dias, Residential energy consumption forecasting using deep learning models, Applied Energy 350 (2023-11) 121705. doi:10.1016/j.apenergy.2023.121705. URL https://linkinghub.elsevier.com/retrieve/pii/S0306261923010693

    2023

  32. [41]

    I. H. Ou Ali, A. Agga, M. Ouassaid, M. Maaroufi, A. Elrashidi, H. Kotb, Predicting short-term energy usage in a smart home using hybrid deep learning models, Front. Energy Res. 12 (2024-09-05) 1323357. doi:10.3389/fenrg.2024.1323357. URL https://www.frontiersin.org/articles/10...

    2024

  33. [42]

    K. Wang, J. Zhang, X. Li, Y. Zhang, Long-term power load forecasting using LSTM-informer with ensemble learning, Electronics 12 (10) (2023-

    2023

  34. [43]

    doi:10.3390/electronics12102175

    2175, number: 10 Publisher = Multidisciplinary Digital Publishing Institute. doi:10.3390/electronics12102175. URL https://www.mdpi.com/2079-9292/12/10/2175

    doi

  35. [44]

    H. Wu, J. Xu, J. Wang, M. Long, Autoformer: Decomposition trans- formers with auto-correlation for long-term series forecasting, 35th Con- ference on Neural Information Processing Systems (2021)

    2021

  36. [45]

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, W. Zhang, Informer: Beyond efficient transformer for long sequence time- series forecasting, Proceedings of the AAAI Conference on Arti- ficial Intelligence 35 (12) (2021-05-18) 11106–11115, number: 12. doi:10.1609/aaai.v3...

    2021 doi

  37. [46]

    Arjovsky, S

    M. Arjovsky, S. Chintala, L. Bottou, Wasserstein generative adversar- ial networks, in: International conference on machine learning, PMLR, 2017, pp. 214–223

    2017

  38. [47]

    Gulrajani, F

    I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, A. C. Courville, Improved training of wasserstein gans, Advances in neural information processing systems 30 (2017)

    2017

  39. [48]

    Hochreiter, J

    S. Hochreiter, J. Schmidhuber, Long short-term memory, Neural com- putation 9 (8) (1997) 1735–1780. 44

    1997

  40. [49]

    J. Ho, A. Jain, P. Abbeel, Denoising diffusion probabilistic models, CoRR abs/2006.11239 (2020). arXiv:2006.11239. URL https://arxiv.org/abs/2006.11239

    2020 arXiv

  41. [50]

    Nichol, P

    A. Nichol, P. Dhariwal, Improved denoising diffusion probabilistic mod- els, CoRR abs/2102.09672 (2021). arXiv:2102.09672. URL https://arxiv.org/abs/2102.09672

    2021 arXiv

  42. [51]

    B. R. H. Geurten, R. Kern, E. Braun, M. Egelhaaf, A syntax of hoverfly flight prototypes, THE JOURNAL OF EXPERIMENTAL BIOLOGY 213 (2010) 2461–247515. doi:doi:10.1242/jeb.036079

    2010 doi

  43. [52]

    Braun, B

    E. Braun, B. Geurten, M. Egelhaaf, Identifying prototypical components in behaviour using clustering algorithms, PLoS ONE 5 (2) (2010-02-22) e9361. doi:10.1371/journal.pone.0009361. URL https://dx.plos.org/10.1371/journal.pone.0009361

    2010 doi

  44. [53]

    Gabrielski, U

    J. Gabrielski, U. H¨ ager, A markov chain model for imputation of electricity consumption time series, in: 2023 58th International Uni- versities Power Engineering Conference (UPEC), 2023-08, pp. 1–6. doi:10.1109/UPEC57427.2023.10294910. URL https://ieeexplore.ieee.org/documen...

    2023

  45. [54]

    Hothorn, T

    T. Hothorn, T. Kneib, P. B¨ uhlmann, Conditional Transformation Mod- els, Journal of the Royal Statistical Society Series B: Statistical Method- ology 76 (1) (2014-01) 3–27. doi:10.1111/rssb.12017

    2014 doi

  46. [55]

    Hothorn, L

    T. Hothorn, L. M¨ ost, P. B¨ uhlmann, Most Likely Transforma- tions, Scandinavian Journal of Statistics 45 (1) (2018) 110–134. doi:10.1111/sjos.12291

    2018 doi

  47. [56]

    Papamakarios, T

    G. Papamakarios, T. Pavlakou, I. Murray, Masked Autoregressive Flow for Density Estimation (2018-06-14). arXiv:1705.07057

    2018 arXiv

  48. [57]

    Papamakarios, E

    G. Papamakarios, E. Nalisnick, D. J. Rezende, S. Mohamed, B. Lakshmi- narayanan, Normalizing Flows for Probabilistic Modeling and Inference, Journal of Machine Learning Research 22 (57) (2021) 1–64. URL http://jmlr.org/papers/v22/19-1028.html 45

    2021

  49. [58]

    R. T. Farouki, The Bernstein Polynomial Basis: A Centennial Retro- spective, Computer Aided Geometric Design 29 (6) (2012-08) 379–419. doi:10.1016/j.cagd.2012.03.001

    2012 doi

  50. [59]

    Fischer, A

    D. Fischer, A. Surmann, W. Biener, O. Selinger-Lutz, From res- idential electric load profiles to flexibility profiles – a stochastic bottom-up approach, Energy and Buildings 224 (2020-10-01) 110133. doi:10.1016/j.enbuild.2020.110133. URL https://www.sciencedirect.com/science/...

    2020

  51. [60]

    Hehlert, B

    P. Hehlert, B. Gerhards, N. Popkov, Sylas KI synthetic powerprofiles OpenMeter (2024). doi:10.25625/A5VZA9. URL https://doi.org/10.25625/A5VZA9

    2024 doi

  52. [61]

    L. Li, Y. Fan, M. Tse, K.-Y. Lin, A review of applications in federated learning, Computers & Industrial Engineering 149 (2020-11-01) 106854. doi:10.1016/j.cie.2020.106854. URL https://www.sciencedirect.com/science/article/pii/S0360835220305532

    2020

  53. [62]

    P. M. Mammen, Federated learning: Opportunities and challenges (2021-01-14). arXiv:2101.05428 [cs], doi:10.48550/arXiv.2101.05428. URL http://arxiv.org/abs/2101.05428

    2021 doi arXiv

  54. [63]

    C. D. Manning, P. Raghavan, H. Sch¨ utze, Introduction to information retrieval, Cambridge university press, 2008

    2008

  55. [64]

    C. Guo, F. Berkhahn, Entity embeddings of categorical variables, arXiv preprint arXiv:1604.06737 (2016)

    2016 arXiv

  56. [65]

    Goodfellow, J

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

    2020

  57. [66]

    J.-Y. Zhu, T. Park, P. Isola, A. A. Efros, Unpaired image-to-image trans- lation using cycle-consistent adversarial networks, in: Proceedings of the IEEE international conference on computer vision, 2017, pp. 2223–2232

    2017

  58. [67]

    Lin, C.-H

    E. Lin, C.-H. Lin, H.-Y. Lane, Relevant applications of generative ad- versarial networks in drug design and discovery: molecular de novo de- sign, dimensionality reduction, and de novo peptide and protein design, Molecules 25 (14) (2020) 3250. 46

    2020

  59. [68]

    Villani, et al., Optimal transport: old and new, Vol

    C. Villani, et al., Optimal transport: old and new, Vol. 338, Springer, 2008

    2008

  60. [69]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international con- ference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, Spri...

    2015

  61. [70]

    D. P. Kingma, P. Dhariwal, Glow: Generative Flow with In- vertible 1x1 Convolutions (2018-07-10). arXiv:1807.03039, doi:10.48550/arXiv.1807.03039

    2018 doi arXiv

  62. [71]

    Van Den Oord, S

    A. Van Den Oord, S. Dieleman, H. Zen, K. Simonyan, O. Vinyals, A. Graves, N. Kalchbrenner, A. Senior, K. Kavukcuoglu, et al., Wavenet: A generative model for raw audio, arXiv preprint arXiv:1609.03499 12 (2016)

    2016 arXiv

  63. [72]

    D. J. Rezende, S. Mohamed, Variational Inference with Normalizing Flows (2016-06-14). arXiv:1505.05770

    2016 arXiv

  64. [73]

    Kobyzev, S

    I. Kobyzev, S. J. Prince, M. A. Brubaker, Normalizing Flows: An In- troduction and Review of Current Methods, IEEE Transactions on Pat- tern Analysis and Machine Intelligence 43 (11) (2021-11) 3964–3979. arXiv:1908.09257, doi:10.1109/TPAMI.2020.2992934

    2021 arXiv

  65. [74]

    Kneib, A

    T. Kneib, A. Silbersdorff, B. S¨ afken, Rage Against the Mean – A Review of Distributional Regression Approaches, Econometrics and Statistics (2021-08-10). doi:10.1016/j.ecosta.2021.07.006

    2021 doi

  66. [75]

    P. F. Baumann, T. Hothorn, D. R¨ ugamer, Deep conditional transforma- tion models, in: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, Springer, 2021, pp. 3–18

    2021

  67. [76]

    B. Sick, T. Hothorn, O. D¨ urr, Deep Transformation Models: Tackling Complex Regression Problems with Neural Network Based Transforma- tion Models, in: 2020 25th International Conference on Pattern Recog- nition (ICPR), IEEE, IEEE, 2021-01, pp. 2476–2481. doi:10/gpd2v4

    2020

  68. [77]

    Arpogaus, M

    M. Arpogaus, M. Voss, B. Sick, M. Nigge-Uricher, O. D¨ urr, Short-Term Density Forecasting of Low-Voltage Load Using Bernstein-Polynomial 47 Normalizing Flows, IEEE Transactions on Smart Grid 14 (6) (2023-11) 4902–4911. arXiv:2204.13939, doi:10.1109/tsg.2023.3254890

    2023 arXiv

  69. [78]

    L. Kook, L. Herzog, T. Hothorn, O. D¨ urr, B. Sick, Deep and interpretable regression models for ordinal outcomes (2021-04-20). arXiv:2010.08376, doi:10.48550/arXiv.2010.08376

    2021 doi arXiv

  70. [79]

    Rugamer, P

    D. Rugamer, P. F. M. Baumann, T. Kneib, T. Hothorn, Probabilistic time series forecasts with autoregressive transformation models, Statis- tics and Computing 33 (2) (2023-02-04) 37. doi:10.1007/s11222-023- 10212-8

    2023 doi

  71. [80]

    D. P. Kingma, J. Ba, Adam: A Method for Stochastic Optimization (2017-01-29). arXiv:1412.6980

    2017 arXiv

  72. [81]

    Germain, K

    M. Germain, K. Gregor, I. Murray, H. Larochelle, MADE: Masked Au- toencoder for Distribution Estimation (2015-06-05). arXiv:1502.03509

    2015 arXiv

  73. [82]

    B. Sick, O. D¨ urr, Interpretable Neural Causal Models with TRAM- DAGs (2025-03-20). arXiv:2503.16206, doi:10.48550/arXiv.2503.16206

    2025 doi arXiv

  74. [83]

    Schreiber, Pomegranate: fast and flexible probabilistic modeling in python, Journal of Machine Learning Research 18 (164) (2018) 1–6

    J. Schreiber, Pomegranate: fast and flexible probabilistic modeling in python, Journal of Machine Learning Research 18 (164) (2018) 1–6. 48

    2018

  75. [84]

    Metric OM WGAN DDPM HMM MABF SLP ±SD sig

    Supplementary Figures Figure S1: Simple statistical metrics -Violin-plots for simple statistical analysis (see also Table 1) of power consumption values for each consumer from the respective ML model compared to the OM data. Metric OM WGAN DDPM HMM MABF SLP ±SD sig. n.s. *** *...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.