Pith. sign in

REVIEW 3 major objections 5 minor 41 references

Tube Loss based Deep Networks For Improving the Probabilistic Forecasting of Wind Speed

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Tube loss gives narrower wind-speed forecast bands at 95% coverage

desk verdict Incremental but useful: applies the authors' Tube loss to wind speed forecasting; the empirical story is plausible, but the theoretical guarantee is overclaimed and the experiments lack error bars and tuning details. read the letter →

arxiv 2505.18284 v1 pith:Y7B2YU2X submitted 2025-05-23 cs.LG cs.AI

Pritam Anand , Aadesh Minz , Asish Joel This is my paper · ORCID
classification cs.LGcs.AI
keywords windspeedforecastingprobabilisticpredictionintervalsTubelossdeeplearningLSTMGRUtemporalconvolutionalnetworks
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

Wind-speed forecasts are more useful when they come with a prediction interval that is both correctly calibrated and narrow. This paper argues that a single loss function, the Tube loss, accomplishes this for deep sequence models without any distributional assumption. The authors train LSTM, GRU, and TCN networks by minimizing the Tube loss plus a width penalty, and report that the resulting intervals hit the target 95% coverage on three hourly wind-speed datasets while being narrower than intervals from quantile regression, QD loss, DeepAR, MDN, and TimeGPT. The practical attraction is a simpler pipeline: one network instead of two separately trained quantile models, a differentiable loss, and intervals that are tighter at the same nominal coverage.

What carries the argument

The central object is the Tube loss (Eq. 1-2), a piecewise linear loss defined over the four regions determined by the lower bound, the upper bound, and a convex combination $r\mu_1 + (1-r)\mu_2$ of the two bounds. Its minimizer asymptotically covers $1-\alpha$ of future observations (the guarantee is proved for the unpenalized loss in [32]), and the parameter $r \in (0,1)$ shifts the tube up or down. The paper's practical addition is the objective in Eq. (3), which adds $\delta \sum |\mu_1 - \mu_2|$ to the Tube loss, and a heuristic that starts at $\delta=0$, checks validation PICP and MPIW, and increments $\delta$ by 0.01 until the validation PICP no longer exceeds the 0.95 target.

What would settle it

Train a Tube-loss model on a synthetic heteroscedastic time series whose conditional quantiles are known, sweep delta from 0 upward, and measure test-set PICP: if PICP falls distinctly below the target 1-α as delta grows, the penalized objective does not preserve the coverage guarantee.

Watch

Extended reading notes

Core claim

The paper's central claim is that Tube-loss-based deep forecasting models outperform recently developed probabilistic wind-speed forecasting baselines in both reliability and sharpness. On the Jaisalmer, Los Angeles, and San Francisco datasets, Tube-loss models achieve a test-set PICP at or above the 0.95 target in every configuration, and they occupy the top three average ranks across the three datasets (TCN+Tube, GRU+Tube, LSTM+Tube). The reported mean interval widths are 2% to 45% smaller than those of the baselines, depending on dataset and competitor. The authors attribute this to the Tube loss's explicit width minimization, its differentiable form, and a delta-recalibration heuristic that widens the interval until validation coverage drops to the target; they also note that the r parameter lets the tube shift toward denser regions when noise is asymmetric.

Load-bearing premise

The load-bearing assumption is that adding the delta width penalty to the Tube loss preserves the asymptotic 95% coverage guarantee that holds for the unpenalized loss; the paper imports the guarantee from [32] but does not prove or cite a version for the penalized objective.

Editorial extensions

If this is right

  • A wind-farm operator can obtain 95% prediction intervals from a single trained network instead of two quantile models, reducing training and tuning overhead.
  • The loss is architecture-agnostic, so the same recipe can ride on any autoregressive deep sequence model, not only LSTM, GRU, and TCN.
  • Coverage stays at the target without inflating width, which makes the intervals directly usable for reserve setting, unit commitment, and market trading decisions.
  • No parametric assumption about wind-speed distribution is needed, so the method applies to asymmetric and heavy-tailed wind regimes.

Reading between the lines

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

  • The delta-recalibration procedure is a validation-set coverage controller; a formal coverage-constrained optimizer could likely replace it and eliminate the gap between the asymptotic guarantee for the unpenalized loss and the penalized objective actually fitted.
  • The paper does not tune the r parameter; since r is claimed to move the tube toward denser regions under skew, data-dependent r selection is a natural extension that could sharpen the intervals further on skewed wind datasets.
  • Because the loss is distribution-free and model-agnostic, the same approach could transfer to solar irradiance or load forecasting, though that transfer is not studied here.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes using the Tube loss of Anand et al. [32] for probabilistic wind speed forecasting within deep autoregressive architectures (LSTM, GRU, TCN). The objective in Eq. (3) adds a width penalty delta * sum |mu1(zi) - mu2(zi)| to the Tube loss, and a validation-based heuristic increments delta by 0.01 until the validation PICP is deemed 'significantly higher' than the 0.95 target. The method is evaluated on three hourly wind speed datasets (Jaisalmer, Los Angeles, San Francisco) against Quantile regression, QD loss, DeepAR, MDN, and TimeGPT, using PICP and MPIW. The paper reports that Tube-loss models meet or exceed the 0.95 coverage while producing narrower intervals than the baselines, and it concludes that Tube-loss models consistently achieve the target calibration across all scenarios.

Significance. If the reported results are reliable, the proposal offers a simple, model-agnostic way to obtain calibrated prediction intervals with deep sequential architectures by training a single network with two outputs. The method is conceptually attractive: it avoids estimating two separate quantile models, is differentiable, and explicitly trades off width against calibration via delta. The paper also provides a straightforward heuristic for setting delta and includes comparisons against several strong baselines. However, the central theoretical claim is not established: the width penalty in Eq. (3) alters the population objective, so the imported asymptotic coverage guarantee from [32] does not automatically hold for the penalized minimizer. In addition, the empirical support is weakened by single-run results with no error bars, missing delta values, and underspecified experimental details. The significance of the work as presented thus depends on a theoretical transfer that is not proven and on experiments that are not fully reproducible.

major comments (3)
  1. [Section II and Eq. (3)] The asymptotic coverage guarantee is imported from [32] for the minimizer of the unpenalized Tube loss, but Eq. (3) minimizes the Tube loss plus delta times the sum of |mu1(zi) - mu2(zi)|. The delta term contributes -delta to the mu1 subgradient and +delta to the mu2 subgradient, so the population minimizer of Eq. (3) does not, in general, satisfy the coverage condition F(mu2) - F(mu1) = 1 - alpha that underpins the result in [32]. The paper needs either a proof that the penalized minimizer retains the asymptotic coverage (with conditions on delta and r), or an explicit statement that calibration is an empirical property achieved by the validation-based delta heuristic. As written, the conclusion that 'Tube loss-based models consistently achieve the target coverage of 0.95 across all scenarios' is not supported by the cited theory.
  2. [Section IV.B, Tables I-III and Table V] All PICP and MPIW results are single-run values, with no standard deviations, seeds, confidence intervals, or number of repetitions. Several key comparisons are very close; for example, on Los Angeles, TCN+Tube has PICP=0.9502 and MPIW=3.555 while TCN+QD has PICP=0.9507 and MPIW=3.548. Moreover, the Los Angeles ranking in Table V places MDN first, and Table VIII shows that MDN's average MPIW is 4.014% better than the Tube-loss average on that dataset. Without error bars or multiple runs, the reported differences cannot be distinguished from randomness, so the central claim of systematic improvement over baselines is not established.
  3. [Section III, recalibration heuristic] The delta-tuning procedure is underspecified: 'significantly higher' is not defined, no statistical test or threshold is given, and there is no rule for decreasing delta when validation PICP falls below 0.95. Since the heuristic directly adjusts delta until the validation PICP is at the target, the test-set PICP values are not an independent confirmation of calibration. The chosen delta values are never reported, so the experiments are not reproducible. This is a load-bearing part of the method because the claimed calibration of the tuned models rests on this heuristic rather than on the imported asymptotic result.
minor comments (5)
  1. [Section III and Eq. (4)] There are several typos: Eq. (4) has ':= =' instead of '=', and the text writes 'PCIP' instead of 'PICP' in Section IV.A; additionally, Section III describes the output bounds as 'mu1(zi) and mu1(zi)' where the second should be 'mu2(zi)'.
  2. [Section III, experimental setup] The paper does not state how the lag window p is selected for each dataset, nor does it report the network hyperparameters (hidden units, learning rate, epochs, batch size) for either the proposed models or the baselines; these details are necessary for reproducibility.
  3. [Section II and Section III] The values of the Tube loss parameter r are never reported for any experiment, and the abstract and text use inconsistent spellings ('San Fransico' vs. 'San Francisco'); these should be corrected.
  4. [Section IV.A] The paper excludes CRPS as 'not relevant' for Tube-loss models, but an interval score or weighted interval score could be computed uniformly for all methods, and CRPS can be computed for baselines that output full predictive distributions; providing such a metric would strengthen the comparison.
  5. [Section IV.A and Table I] TimeGPT is used as a pretrained foundation model with no fine-tuning, but this is not stated explicitly; the comparison would be fairer if the paper clarified the training protocol used for each baseline.

Circularity Check

2 steps flagged · score 4.0 of 10

The asymptotic coverage guarantee is imported from the authors' own Tube loss paper [32] for the unpenalized objective, while the implemented Eq. (3) adds a width penalty that invalidates the transfer; the reported 0.95 coverage is then produced by the validation-based delta heuristic.

  1. self citation load bearing [Section II (after Eq. 2) and Section III, Eq. (3)]
    "The coverage obtained by PI with minimizer of the Tube loss function (ˆµ1(x), ˆµ2(x)) reaches to the target 1 − α asymptotically. The proof of this is is detailed in [32]. ... The Tube loss based deep forecasting architecture ... minimize the following problem ... ρr 1−α(xi+1, µ1(zi), µ2(zi)) + δ Σ (|µ1(zi)−µ2(zi)|) (3)"

    The only cited support for the 1−α coverage guarantee is [32], a prior preprint by the same first author, and that proof is for the unpenalized Tube loss. The actual method minimizes Eq. (3), which appends the width penalty δΣ|μ1−μ2|. This penalty changes the first-order conditions: it contributes −δ to the μ1 subgradient and +δ to the μ2 subgradient, so the population minimizer no longer satisfies the quantile-spacing condition F(μ2)−F(μ1)=1−α that underlies the unpenalized theorem. No proof or external citation is supplied for the penalized objective. Thus the theoretical calibration claim for the proposed models is inherited by self-citation and does not apply to the fitted objective.

  2. fitted input called prediction [Section III, delta-tuning heuristic, and Section IV-D Conclusion]
    "If the observed PICP on the validation set is significantly higher than the target calibration level of 0.95, there is an opportunity to reduce the MPIW value, since MPIW increases with PICP. In such cases, we increment the value of δ by 0.01 and retrain our deep forecasting architecture to achieve lower MPIW values. ... Tube loss-based models consistently achieve the target coverage of 0.95 across all scenarios, regardless of the underlying deep architecture."

    The δ parameter is tuned directly against the target coverage: whenever validation PICP is 'significantly higher' than 0.95, δ is increased and the model retrained until validation coverage is brought down toward the target. The concluding claim of consistent 0.95 test coverage is therefore the product of this validation recalibration loop, not of the imported asymptotic guarantee. No δ values, significance thresholds, seeds, or repeated runs are reported, so the test-set PICP values near 0.95 cannot be separated from the validation-fitting procedure. This is a fitted parameter presented as an intrinsic property of the Tube-loss method.

full rationale

The empirical ranking of Tube+LSTM/GRU/TCN against Quantile, QD, DeepAR, MDN, and TimeGPT (Tables I–III and VII) is a direct external comparison on three wind datasets, so the central empirical claim has independent content and is not circular. The circularity is confined to the theoretical and calibration narrative. The paper advertises asymptotic coverage guarantees for its models but imports the guarantee from [32], a preprint whose first author overlaps with this paper, and that guarantee applies to the unpenalized Tube loss only. The implemented objective in Eq. (3) adds the width penalty δΣ|μ1−μ2|, which changes the population minimizer and therefore invalidates the imported theorem; no new proof is given. The observed 0.95 coverage is then obtained by the Section III heuristic that increments δ whenever validation PICP exceeds 0.95, meaning the target calibration is tuned into the model rather than derived from the cited theory. Because the benchmark comparisons are external and the Tube loss itself is a real prior proposal, the paper is not wholly circular, but the load-bearing theoretical self-citation and the validation-driven calibration justify a moderate score of 4.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim depends on two free hyperparameters (δ and, to a lesser extent, r) inherited from the Tube loss framework, plus unreported architecture and data choices. The coverage guarantee is imported from the authors' own prior paper and is not re-derived for the modified objective. No new physical or model entities are introduced.

free parameters (4)
  • δ (width penalty weight in Eq. 3) = not reported
    The δ parameter trades calibration against PI width. The paper tunes δ on the validation set (increment by 0.01 from 0) to keep PICP near 0.95 and reduce MPIW, but reports no chosen δ values per model or dataset. This data-fitted hyperparameter directly affects the reported MPIW improvements.
  • r (Tube loss skew parameter) = 0.5 (default)
    The r parameter controls the position of the PI tube and is described as tunable for asymmetric noise, but the experiments use the default value 0.5. It is a user-chosen constant inherited from [32].
  • Lag window p (autoregressive order) = not reported
    The paper states that 'proper choice of the lag p is obtained or tuned' but does not give the values used for each dataset. This hyperparameter shapes the input features for all models.
  • Network hyperparameters (hidden units, learning rate, epochs, batch size) = not reported
    Training details for LSTM, GRU, TCN, and all baselines are not provided, making it impossible to assess whether architectures were compared fairly or to reproduce the results.
assumptions (3)
  • domain assumption The Tube loss minimizer asymptotically achieves target calibration 1−α (from [32]).
    The paper relies on this theoretical guarantee, citing [32] by the same first author, to justify the calibration of the method. The guarantee is for the unpenalized Tube loss minimizer over function classes, not for a deep network trained with SGD.
  • ad hoc to paper The δ-penalized objective (Eq. 3) preserves the asymptotic coverage guarantee of the unpenalized Tube loss.
    The paper adds the width penalty δ without proving or citing a result that the penalized minimizer still covers 1−α of future observations. The empirical PICP near 0.95 is obtained via validation tuning, not by a theoretical guarantee for the modified objective.
  • domain assumption The conditional distribution (x_{i+1} | z_i) is stationary enough over the test horizon for the auto-regressive transfer to hold.
    The paper trains on the first 70% of each time series and tests on the last 30%, assuming the learned relationship generalizes. No stationarity or drift checks are provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tube Loss based Deep Networks For Improving the Probabilistic Forecasting of Wind Speed." pith.science (2026). https://pith.science/paper/Y7B2YU2X

@misc{pith2026250518284,
  author       = {Pith},
  title        = {Pith review of: Tube Loss based Deep Networks For Improving the Probabilistic Forecasting of Wind Speed},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y7B2YU2X}},
  note         = {Machine review of arXiv:2505.18284}
}
abstract

Uncertainty Quantification (UQ) in wind speed forecasting is a critical challenge in wind power production due to the inherently volatile nature of wind. By quantifying the associated risks and returns, UQ supports more effective decision-making for grid operations and participation in the electricity market. In this paper, we design a sequence of deep learning based probabilistic forecasting methods by using the Tube loss function for wind speed forecasting. The Tube loss function is a simple and model agnostic Prediction Interval (PI) estimation approach and can obtain the narrow PI with asymptotical coverage guarantees without any distribution assumption. Our deep probabilistic forecasting models effectively incorporate popular architectures such as LSTM, GRU, and TCN within the Tube loss framework. We further design a simple yet effective heuristic for tuning the $\delta$ parameter of the Tube loss function so that our deep forecasting models obtain the narrower PI without compromising its calibration ability. We have considered three wind datasets, containing the hourly recording of the wind speed, collected from three distinct location namely Jaisalmer, Los Angeles and San Fransico. Our numerical results demonstrate that the proposed deep forecasting models produce more reliable and narrower PIs compared to recently developed probabilistic wind forecasting methods.

Figures

Figures reproduced from arXiv: 2505.18284 by the authors.

Figure 1
Figure 1. Tube loss function for 1 − α = 0.9 u1 := y−µ1(x) and u2 := y−µ2(x) in the Tube loss function (1), which reduces it to ρ r 1−α(y, µ1(x), µ2(x)) =    t(y − µ2(x)), if y > µ2(x). (1 − t)(µ2(x) − y), if µ1(x) ≤ y ≤ µ2(x) and y ≥ rµ2(x) + (1 − r)µ1(x). (1 − t)(y − µ1(x)), if µ1(x) ≤ y ≤ µ2(x) and y < rµ2(x) + (1 − r)µ1(x). t(µ1(x) − y), if y < µ1(x). (2) Let us denote the (ˆµ1(x), µˆ2(x)) is the minimi… view at source ↗
Figure 2
Figure 2. Tube loss methodology: red line represents the convex combination [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Flow Diagram: Probabilistic forecasting with Tube loss based deep learning models [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Average Improvement obtained by the Tube loss model over existing [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Plot of GRU-Tube on Jaisalmer Dataset loss based deep forecasting method can obtain the significant improvement of the MPIW values as compared to the other existing model. REFERENCES [1] Global Wind Energy Council. Gwec global wind report 2019. Global wind energy counc…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 37 canonical work pages

  1. [32]

    Tube Loss: A Novel Approach for Prediction Interval Estimation

    Pritam Anand, Tathagata Bandyopadhyay, and Suresh Chandra. Tube loss: A novel approach for prediction interval estimation and proba- bilistic forecasting. arXiv preprint arXiv:2412.06853 , 2024

    work page Pith review arXiv 2024

  2. [1]

    Gwec global wind report 2019

    Global Wind Energy Council. Gwec global wind report 2019. Global wind energy council: Bonn, Germany , 2017

    work page 2019

  3. [2]

    Probabilistic and deterministic wind speed forecasting based on non- parametric approaches and wind characteristics information

    Jiani Heng, Yongmiao Hong, Jianming Hu, and Shouyang Wang. Probabilistic and deterministic wind speed forecasting based on non- parametric approaches and wind characteristics information. Applied Energy, 306:118029, 2022

    work page 2022

  4. [3]

    Advanced deep learning approach for probabilis- tic wind speed forecasting

    Mousa Afrasiabi, Mohammad Mohammadi, Mohammad Rastegar, and Shahabodin Afrasiabi. Advanced deep learning approach for probabilis- tic wind speed forecasting. IEEE Transactions on Industrial Informatics , 17(1):720–727, 2020

    work page 2020

  5. [4]

    A multi-model combination approach for probabilistic wind power forecasting

    You Lin, Ming Yang, Can Wan, Jianhui Wang, and Yonghua Song. A multi-model combination approach for probabilistic wind power forecasting. IEEE Transactions on Sustainable Energy , 10(1):226–237, 2018

    work page 2018

  6. [5]

    A multi- state model for exploiting the reserve capability of wind power

    You Lin, Yi Ding, Yonghua Song, and Chuangxin Guo. A multi- state model for exploiting the reserve capability of wind power. IEEE Transactions on Power Systems , 33(3):3358–3372, 2017

    work page 2017

  7. [6]

    Expected value and chance constrained stochastic unit commitment ensuring wind power utilization

    Chaoyue Zhao, Qianfan Wang, Jianhui Wang, and Yongpei Guan. Expected value and chance constrained stochastic unit commitment ensuring wind power utilization. IEEE Transactions on Power Systems , 29(6):2696–2705, 2014

    work page 2014

  8. [7]

    Pareto optimal prediction intervals of electricity price

    Can Wan, Ming Niu, Yonghua Song, and Zhao Xu. Pareto optimal prediction intervals of electricity price. IEEE Transactions on Power Systems, 32(1):817–819, 2016

    work page 2016

Show all 41 references
  1. [8]

    Quantile regression, volume 38

    Roger Koenker. Quantile regression, volume 38. Cambridge university press, 2005

    2005

  2. [9]

    Wind farm power uncertainty quantification using a mean- variance estimation method

    Abbas Khosravi, Saeid Nahavandi, Douglas Creighton, and Jafreezal Jaafar. Wind farm power uncertainty quantification using a mean- variance estimation method. In 2012 IEEE International Conference on Power System Technology (POWERCON) , pages 1–6. IEEE, 2012

    2012

  3. [10]

    An optimized mean variance es- timation method for uncertainty quantification of wind power forecasts

    Abbas Khosravi and Saeid Nahavandi. An optimized mean variance es- timation method for uncertainty quantification of wind power forecasts. International Journal of Electrical Power & Energy Systems , 61:446– 454, 2014

    2014

  4. [11]

    Sparse online warped gaussian process for wind power probabilistic forecasting

    Peng Kou, Feng Gao, and Xiaohong Guan. Sparse online warped gaussian process for wind power probabilistic forecasting. Applied energy, 108:410–428, 2013

    2013

  5. [12]

    Probabilistic wind speed forecast- ing on a grid based on ensemble model output statistics

    Michael Scheuerer and David M ¨oller. Probabilistic wind speed forecast- ing on a grid based on ensemble model output statistics. 2015

    2015

  6. [13]

    Wind- friendly flexible ramping product design in multi-timescale power sys- tem operations

    Mingjian Cui, Jie Zhang, Hongyu Wu, and Bri-Mathias Hodge. Wind- friendly flexible ramping product design in multi-timescale power sys- tem operations. IEEE Transactions on Sustainable Energy , 8(3):1064– 1075, 2017

    2017

  7. [14]

    A novel wind power probabilistic forecasting approach based on joint quantile regression and multi-objective optimization

    Jianming Hu, Jingwei Tang, and Yingying Lin. A novel wind power probabilistic forecasting approach based on joint quantile regression and multi-objective optimization. Renewable Energy, 149:141–164, 2020

    2020

  8. [15]

    Direct quantile regression for nonparametric probabilistic forecasting of wind power generation

    Can Wan, Jin Lin, Jianhui Wang, Yonghua Song, and Zhao Yang Dong. Direct quantile regression for nonparametric probabilistic forecasting of wind power generation. IEEE Transactions on Power Systems , 32(4):2767–2778, 2016

    2016

  9. [16]

    A regional wind power probabilistic forecast method based on deep quantile regression

    Yixiao Yu, Ming Yang, Xueshan Han, Yumin Zhang, and Pingfeng Ye. A regional wind power probabilistic forecast method based on deep quantile regression. IEEE Transactions on Industry Applications , 57(5):4420–4427, 2021

    2021

  10. [17]

    Wind power probabilistic forecasting based on combined decomposition and deep learning quantile regression

    Zhenglin Zhu, Yusen Xu, Junzhao Wu, Yiwen Liu, Jianwei Guo, and Haixiang Zang. Wind power probabilistic forecasting based on combined decomposition and deep learning quantile regression. Frontiers in Energy Research, 10:937240, 2022

    2022

  11. [18]

    Ensemble deep learning- based non-crossing quantile regression for nonparametric probabilistic forecasting of wind power generation

    Wenkang Cui, Can Wan, and Yonghua Song. Ensemble deep learning- based non-crossing quantile regression for nonparametric probabilistic forecasting of wind power generation. IEEE Transactions on Power Systems, 38(4):3163–3178, 2022

    2022

  12. [19]

    A large-scale multi-objective evolutionary quantile estimation model for wind power probabilistic forecasting

    Jianhua Zhu and Yaoyao He. A large-scale multi-objective evolutionary quantile estimation model for wind power probabilistic forecasting. IEEE Transactions on Evolutionary Computation , 2024

    2024

  13. [20]

    Lower upper bound estimation method for construction of neural network-based prediction intervals

    Abbas Khosravi, Saeid Nahavandi, Doug Creighton, and Amir F Atiya. Lower upper bound estimation method for construction of neural network-based prediction intervals. IEEE transactions on neural networks, 22(3):337–346, 2010

    2010

  14. [21]

    Probabilistic wind power forecasts considering different nwp models

    Sheng-Hong Wu and Yuan-Kang Wu. Probabilistic wind power forecasts considering different nwp models. In 2020 International Symposium on Computer , Consumer and Control (IS3C) , pages 428–431. IEEE, 2020

    2020

  15. [22]

    An evolutionary multiobjective knee-based lower upper bound estima- tion method for wind speed interval forecast

    Kaiwen Li, Tao Zhang, Rui Wang, Ling Wang, and Hisao Ishibuchi. An evolutionary multiobjective knee-based lower upper bound estima- tion method for wind speed interval forecast. IEEE Transactions on Evolutionary Computation , 26(5):1030–1042, 2021

    2021

  16. [23]

    Probabilistic wind power forecasting based on spiking neural network

    Huaizhi Wang, Wenli Xue, Yitao Liu, Jianchun Peng, and Hui Jiang. Probabilistic wind power forecasting based on spiking neural network. Energy, 196:117072, 2020

    2020

  17. [24]

    An intelligent deep learning based prediction model for wind power generation

    Abdulaziz Almutairi and Omar Alrumayh. An intelligent deep learning based prediction model for wind power generation. Computers and Electrical Engineering , 101:108000, 2022

    2022

  18. [25]

    A novel interval estimation framework for wind power forecasting using multi- objective gradient descent optimization

    Yinsong Chen, Samson Yu, Chee Peng Lim, and Peng Shi. A novel interval estimation framework for wind power forecasting using multi- objective gradient descent optimization. Sustainable Energy, Grids and Networks, 38:101363, 2024

    2024

  19. [26]

    A new lower and upper bound estimation model using gradient descend training method for wind speed interval prediction

    Fangjie Liu, Chaoshun Li, Yanhe Xu, Geng Tang, and Yuying Xie. A new lower and upper bound estimation model using gradient descend training method for wind speed interval prediction. Wind Energy , 24(3):290–304, 2021

    2021

  20. [27]

    High-quality prediction intervals for deep learning: A distribution-free, ensembled approach

    Tim Pearce, Alexandra Brintrup, Mohamed Zaki, and Andy Neely. High-quality prediction intervals for deep learning: A distribution-free, ensembled approach. In International conference on machine learning , pages 4075–4084. PMLR, 2018

    2018

  21. [28]

    A new wind power interval prediction approach based on reservoir computing and a quality-driven loss function

    Jianming Hu, Yingying Lin, Jingwei Tang, and Jing Zhao. A new wind power interval prediction approach based on reservoir computing and a quality-driven loss function. Applied Soft Computing , 92:106327, 2020

    2020

  22. [29]

    On the properties of neural machine translation: Encoder- decoder approaches

    Kyunghyun Cho, Bart Van Merri ¨enboer, Dzmitry Bahdanau, and Yoshua Bengio. On the properties of neural machine translation: Encoder- decoder approaches. arXiv preprint arXiv:1409.1259 , 2014

    2014 arXiv

  23. [30]

    Long short-term memory

    Sepp Hochreiter and J ¨urgen Schmidhuber. Long short-term memory. Neural computation , 9(8):1735–1780, 1997

    1997

  24. [31]

    Temporal convolutional networks for action segmentation and detection

    Colin Lea, Michael D Flynn, Rene Vidal, Austin Reiter, and Gregory D Hager. Temporal convolutional networks for action segmentation and detection. In proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 156–165, 2017

    2017

  25. [33]

    Deepar: Probabilistic forecasting with autoregressive recurrent networks

    David Salinas, Valentin Flunkert, Jan Gasthaus, and Tim Januschowski. Deepar: Probabilistic forecasting with autoregressive recurrent networks. International journal of forecasting , 36(3):1181–1191, 2020

    2020

  26. [34]

    Mixture density networks

    Christopher M Bishop. Mixture density networks. 1994. 9

    1994

  27. [35]

    Timegpt-

    Azul Garza, Cristian Challu, and Max Mergenthaler-Canseco. Timegpt-

  28. [36]

    arXiv preprint arXiv:2310.03589 , 2023

    2023 arXiv

  29. [37]

    Short-term wind speed interval prediction using improved quality-driven loss based gated multi- scale convolutional sequence model

    Adnan Saeed, Chaoshun Li, and Zhenhao Gan. Short-term wind speed interval prediction using improved quality-driven loss based gated multi- scale convolutional sequence model. Energy, 300:131590, 2024

    2024

  30. [38]

    Probabilistic wind power forecasting using optimized deep auto-regressive recur- rent neural networks

    Parul Arora, Seyed Mohammad Jafar Jalali, Sajad Ahmadian, Bijaya K Panigrahi, Ponnuthurai N Suganthan, and Abbas Khosravi. Probabilistic wind power forecasting using optimized deep auto-regressive recur- rent neural networks. IEEE Transactions on Industrial Informatics , 19(3)...

    2022

  31. [39]

    An improved mixture density network via wasserstein distance based adversarial learning for probabilistic wind speed predictions

    Luoxiao Yang, Zhong Zheng, and Zijun Zhang. An improved mixture density network via wasserstein distance based adversarial learning for probabilistic wind speed predictions. IEEE Transactions on Sustainable Energy, 13(2):755–766, 2021

    2021

  32. [40]

    Improved deep mixture density network for regional wind power probabilistic forecasting

    Hao Zhang, Yongqian Liu, Jie Yan, Shuang Han, Li Li, and Quan Long. Improved deep mixture density network for regional wind power probabilistic forecasting. IEEE Transactions on Power Systems , 35(4):2549–2560, 2020

    2020

  33. [41]

    Short-term wind speed and power forecasting using an ensemble of mixture density neural networks

    Zhongxian Men, Eugene Yee, Fue-Sang Lien, Deyong Wen, and Yong- sheng Chen. Short-term wind speed and power forecasting using an ensemble of mixture density neural networks. Renewable Energy , 87:203–211, 2016

    2016

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.