Pith. sign in

REVIEW 4 major objections 5 minor 40 references

Context-Aware Conformal Prediction recalibrates black-box day-ahead solar forecasts by weighting past errors by how similar their context is to the current forecast, yielding materially narrower prediction intervals that still hit target co

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 09:19 UTC pith:WMPITZEB

load-bearing objection Useful applied conformal forecasting paper, but the daily tuning budget CACP gets (and baselines don't) likely explains part of the claimed width gains. the 4 major comments →

arxiv 2510.15780 v2 pith:WMPITZEB submitted 2025-10-17 stat.AP cs.LG

Enhanced Renewable Energy Forecasting using Context-Aware Conformal Prediction

Alireza Moradi , Mathieu Tanneau , Reza Zandehshahvar , Pascal Van Hentenryck This is my paper
classification stat.AP cs.LG MSC 62G1562M10
keywords conformal predictionprobabilistic forecastingsolar power forecastingprediction intervalscalibrationweighted quantileday-ahead forecastingrenewable energy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Solar forecasters often produce intervals that miss their stated coverage; the paper reports raw forecast miscoverage of 17-22% at the system level and 28-52% at individual sites for 90% target intervals in its data. The paper proposes Context-Aware Conformal Prediction (CACP), a post-hoc calibration layer that reweights historical forecast errors by how similar their context is to the current forecast, using time of day, season, recent actual generation, and phase of the solar day as context. In experiments on a year of day-ahead solar forecasts in three U.S. grid regions, CACP keeps empirical coverage at or near the target while cutting interval width relative to standard conformal baselines and the raw forecasts, achieving the lowest Winkler score at every target coverage level. The claim is that this improvement in the reliability-efficiency tradeoff is available without retraining or accessing the underlying forecasting model, and at a per-day cost of a few minutes of CPU time. A sympathetic reader would care because system operators and site owners often cannot retrain third-party forecast models, making calibration-only reliability layers attractive.

Core claim

CACP makes calibration local. For a test point with context vector x_t, each calibration sample tau receives a weight psi(x_t, x_tau); the prediction interval is built from the weighted 1-alpha quantile of the calibration conformity scores (the signed distances of past observations from the base model's interval). The context vector combines recent actual generation, sine-cosine embeddings of hour, day, and month, and the normalized phase of the solar day. Three weight functions are studied: an RBF kernel, k-means cluster membership, and k-nearest neighbors. The paper's experiments show that this local weighting produces intervals that are roughly 30% narrower than CQR's on one of the three

What carries the argument

The weighted conformity quantile is the mechanism that carries the argument. Instead of taking the global quantile of all calibration conformity scores, CACP forms a weighted empirical distribution in which each sample's weight is proportional to psi(x_t, x_tau), a non-negative similarity between the test context and the calibration context; the adjusted interval is then the base interval widened or narrowed by the weighted score quantile. The context covariate x_t is the concatenation of recent actuals, hour/day/month sine-cosine embeddings, and the normalized solar-day phase. This is what converts a global recalibration into a locally adaptive one, and it is what the paper credits for both

Load-bearing premise

The load-bearing premise is that similar context features imply similar forecast-error distributions, and that the preceding week's data is representative of the next day's error regime; if either fails, CACP's local weighting is no better than a seasonal recalibrator and the reported sharpness gains can disappear.

What would settle it

Hold out a period of unusual weather (e.g., widespread cloud or a storm) whose error regime is not captured by hour, season, recent actuals, or solar-day phase, and compare CACP's conditional coverage on those days with CQR's; if CACP's coverage falls materially below target while CQR holds, the similarity premise is false. Concretely, one could take days where the base forecast's error is large despite the context features looking normal, and check whether the weighted quantile still tracks the empirical error.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Black-box and vendor-supplied solar forecasts can be made more reliable by a post-processing layer that uses only historical forecast quantiles, actuals, and timestamps, without retraining or accessing the base model.
  • On the paper's numbers, CACP at 90% coverage reduces average interval width by about 30% relative to CQR on one of the three systems (0.0908 vs 0.1298) while keeping empirical coverage above target, giving operators tighter uncertainty bands for the same promised reliability.
  • Hour-by-hour coverage stays close to the 80% target across all daylight hours, correcting the systematic undercoverage in the morning and evening and the overcoverage at midday that the baselines show.
  • The daily re-tuning and feature-selection loop makes CACP track seasonal and weather-driven drift without a heavy compute budget; this is a direct route to adaptive calibration in operational settings.
  • Because CACP operates on generic conformity scores, the same method applies to any quantile-based forecaster, not just the solar model tested here.

Where Pith is reading between the lines

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

  • A testable extension: compare CACP against a CQR variant that is re-tuned daily on the same seven-day validation window; if the gap narrows, part of the reported gain comes from the larger model-selection budget rather than the weighting itself.
  • The same context-feature template could be applied to wind or load forecasting by replacing solar-day phase with the relevant daily cycle; the paper lists these as future work and nothing in the mechanism prevents it.
  • The hour-of-day analysis implies a diagnostic use: CACP's shifting conformity-score distribution reveals when the base model is systematically overconfident (midday) or underconfident (evening), which an operator could use to flag model weaknesses, not just to build intervals.
  • The conformal guarantee under exchangeability does not automatically transfer to these data-dependent weights, so a deployment should pair CACP with online coverage monitoring rather than assuming fixed coverage.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces Context-Aware Conformal Prediction (CACP), a post-hoc calibration layer for probabilistic day-ahead solar forecasts. CACP computes similarity weights between the test covariates and each calibration sample (via RBF/Laplacian kernels, KMeans, or KNN), forms a weighted quantile of conformity scores (Eq. 9), and dynamically re-tunes hyperparameters and feature subsets on a rolling validation week (Sec. III-C). Experiments on NREL's 2019 datasets for MISO, ERCOT, and SPP, at both site and system level, compare CACP against NREL raw forecasts, CQR, AdaptiveCP, NexCP, and HopCPT. The reported tables show that CACP variants achieve the lowest Winkler scores in essentially every ISO, coverage level, and aggregation setting; for example, MISO system-level 90% WS is 0.1065 for CACP-KNN versus 0.1262 for HopCPT and 0.1386 for CQR, with PICP around 91.2%.

Significance. If the empirical results are taken at face value, the paper offers a practically valuable contribution: a lightweight, model-agnostic calibration layer that improves interval sharpness while maintaining empirical coverage, without retraining the base forecast model. The large-scale evaluation across three U.S. ISOs, two aggregation levels, multiple coverage targets, and several strong baselines is a genuine strength, and the consistency of the WS improvements is striking. However, the central attribution of the gains to the context-aware weighting mechanism is not yet established, because CACP receives a much larger model-selection budget than the baselines, and because the weighted-quantile procedure is presented without a validity condition. With appropriate ablations and uncertainty quantification, this could become a solid applied contribution; in its current form, the evidence supports the existence of a useful post-hoc calibration procedure but not the specific mechanism claimed.

major comments (4)
  1. [Sec. III-C, Table I, Sec. IV-A] Unequal model-selection budget confounds the central claim. CACP performs daily hyperparameter optimization and feature selection over a large grid (RBF/Laplacian kernel with gamma in {0.5,1,2}, KMeans K in {3,5,8,12}, KNN k in {50,100,200,500,1000}, plus feature subsets), while the baselines have fixed or near-fixed hyperparameters. Tables II-III therefore compare a heavily tuned method against essentially untuned baselines. The reported gains may reflect the additional tuning budget rather than the similarity-weighting mechanism. Please provide an ablation that fixes CACP's hyperparameters (or limits tuning to the same frequency/budget as the baselines), and/or give the baselines the same daily selection budget. Also report the features and hyperparameters selected over time, and their stability, to demonstrate that the weighting mechanism itself is the driver.
  2. [Sec. III-A, Eqs. (8)-(9)] No coverage guarantee is established for the weighted quantile. In the weighted conformal prediction theory of [2], finite-sample validity requires weights proportional to known likelihood ratios. Here the weights are arbitrary similarity functions of the test covariate and calibration covariates, so the theoretical guarantee does not transfer. The paper should state explicitly that CACP is a heuristic procedure whose coverage is only empirical, or give a specific assumption (e.g., conditional exchangeability given the context features) under which the weighted quantile is calibrated. This matters because the title and the 'reliability' claim imply a conformal guarantee, and readers may otherwise assume the validity result of [2] applies.
  3. [Sec. IV-A, Sec. V-A, Tables II-III] The evaluation lacks uncertainty quantification. All metrics are single point estimates over one year, with no standard errors, confidence intervals, or significance tests across sites or days. Given the strong temporal autocorrelation and the modest differences in some comparisons (e.g., SPP 60% WS 0.0865 for both CACP-KNN and CACP-Kernel), the reader cannot judge whether the reported advantages are statistically reliable. Please report per-site or per-day distributions of the metrics and perform appropriate paired tests, at least for the main WS comparisons.
  4. [Sec. III-C, Sec. IV-A] Potential overlap between tuning and calibration sets. The paper states that 'for each test day, the preceding week serves as the validation set (for parameter tuning), while calibration is performed using all data available prior to that day.' If the calibration set includes the tuning week, then the hyperparameters are selected on a subset of the calibration data, which can bias the empirical coverage estimates. Please clarify whether the validation week is excluded from the calibration set, and if not, rerun the evaluation with the validation week excluded or otherwise account for the overlap.
minor comments (5)
  1. [Sec. III-B1, Eq. (13)] The definition of H^l_{t,k} is ambiguous: the notation 'past k observations with a lag l' and the bracket [y_{t-l}, ..., y_{t-l-k}] implies k+1 entries. Please clarify the indexing.
  2. [Sec. IV-B, Eq. (17)] The Hopfield loss in Eq. (17) is hard to parse: the norm notation and the dimensions of A and s_{1:T} are not defined. Please spell out the matrix/vector dimensions and the meaning of |s_{1:T}|.
  3. [Fig. 4 / Table IV] Inconsistent date: Fig. 4 says '2019-04-20' while Table IV says '2019-04-2'. Please align.
  4. [Sec. III-C] Typo: 'off-the-self' should be 'off-the-shelf'. Also, 'hypertuning' is nonstandard; consider 'hyperparameter tuning'.
  5. [General] No code or data availability statement is provided. For an applied forecasting paper, a link to the code and preprocessed data would substantially aid reproducibility.

Circularity Check

0 steps flagged

No significant circularity; CACP's intervals are out-of-sample weighted quantiles, with only non-load-bearing self-citations.

full rationale

The derivation chain is self-contained. The central object, Eq. (9), defines the conformal adjustment as a weighted quantile of calibration conformity scores, with weights given by Eq. (8). This is the method's definition rather than a hidden reduction: the weights depend on covariates x_t, x_tau, not on the test outcomes y_t. The reported PICP/AIW/WS metrics are evaluated on T_test after daily recalibration; per Sec. IV-A, the preceding week is the validation set for tuning, while calibration uses all data prior to the test day, so the headline reliability/efficiency improvement is an empirical out-of-sample claim, not an identity. The dynamic hypertuning and feature selection in Sec. III-C maximize validation-set Winkler score, but they do not fit the test-day outcomes. The only self-citations, [18] and [39], appear in the introduction as examples of scenario-generation literature and are not load-bearing for the coverage, sharpness, or weighting claims; no uniqueness theorem or ansatz is imported from the authors' prior work. Reference [2] is external and used only for background. The strongest concern raised by the paper's design — that CACP receives a much larger daily hyperparameter/feature-selection budget than the baselines — is a comparison-fairness and attribution issue, not a circularity: it does not make the output equivalent to its inputs by construction. Accordingly, no circular step is exhibited, and the self-citations are too peripheral to raise the score beyond the low end of the normal range.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

CACP introduces no invented entities: the normalized time-of-solar-day variable rho_t (Eq. 15) is a constructed covariate, a re-parametrization of time, not a new physical or theoretical entity. The free parameters are the daily-tuned hyperparameters of the three weighting schemes (K, k, gamma), the chosen kernel type, the selected feature subset, and the unspecified lag/history of the historical-actuals feature. The formal machinery is inherited from weighted conformal prediction [2]; the paper's own contribution is the engineering of context features and the online tuning loop, whose core premise, local stationarity of conformity scores in feature space, is assumed rather than proven.

free parameters (5)
  • RBF/Laplacian kernel bandwidth gamma (and kernel type) = not reported; search space {0.5, 1.0, 2.0}
    Controls the similarity decay in Eq. (10); re-tuned daily on a 7-day validation window by Winkler score; selected value per site/system/day not reported.
  • KMeans cluster count K = not reported; search {3, 5, 8, 12}
    Granularity of the local weighting in Eq. (11); fitted to validation performance.
  • KNN neighbor count k = not reported; search {50, 100, 200, 500, 1000}
    Size of the local calibration neighborhood in Eq. (12); balances variance vs. locality of the weighted quantile.
  • Feature subset selection = selected subset not reported
    Sec. III-C selects the 'most informative subset of features' periodically by validation performance; outcomes undisclosed, so the mechanism's contribution is partly unobservable.
  • Historical-actuals lag l and window k (Eq. 13) = unspecified
    Values of l and k for H^l_{t,k} are never given, so the exact covariate vector in Eq. (16) is under-determined.
axioms (4)
  • standard math Weighted quantile of conformity scores yields calibrated intervals (Barber et al. 2023 framework)
    Section III-A builds directly on [2]'s weighted conformal prediction; the paper explicitly refers the reader to [2] for the theory.
  • domain assumption Covariate similarity in the chosen features implies similarity of conformity-score distributions (local stationarity)
    The load-bearing premise behind Eqs. (8)-(12): without it, re-weighting by feature distance cannot improve conditional calibration. Not guaranteed by CP theory for data-dependent weights.
  • standard math The CQR conformity score in Eq. (5) correctly captures the base model's miscalibration
    The max-based score is the standard CQR construction of Romano et al. [22], used for all methods including baselines.
  • domain assumption NREL PERFORM forecasts/actuals and the sunrise/sunset data are accurate and 2019 is representative
    The empirical evaluation (Tables II-III) and the solar-day variable rho_t (Eq. 15) rely on the data layer; no independent validation of the data is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 13478 in / 21584 out tokens · 184167 ms · 2026-08-04T09:19:35.742389+00:00 · methodology

0 comments
read the original abstract

Artificial intelligence (AI) is increasingly used to support renewable energy forecasting and grid operations. As renewable penetration grows, reliable probabilistic forecasting is becoming essential for managing uncertainty and supporting risk-aware operational decision-making. However, these forecasts often suffer from miscalibration due to temporal variability, changing weather conditions, and heterogeneous operating regimes. In many real-world settings, renewable energy forecasts are provided by external sources, vendors, or independently trained systems, making retraining infeasible because of limited model access or computational constraints. This creates a need for efficient and model-agnostic methods that can improve forecast reliability after they are produced. This paper presents Context-Aware Conformal Prediction (CACP), a framework for calibrating renewable energy forecasts. The proposed method relies on a weighting mechanism during the calibration procedure which assigns higher weights to historical observations that are more similar to the target forecasting condition. This enables adaptive prediction intervals that reflect local uncertainty regimes without requiring access to, or retraining of, the underlying forecasting model. Experiments are performed on a large-scale dataset from National Renewable Energy Laboratory (NREL) day-ahead solar forecasting, covering multiple systems including MISO, ERCTO, and SPP. The results show that CACP improves the reliability-efficiency tradeoff at both site and system levels compared to NREL's base forecasting model and the other conformal prediction baselines. These results suggest that CACP can serve as a practical reliability-enhancement layer for trustworthy AI-enabled renewable energy forecasting and operational decision support.

Figures

Figures reproduced from arXiv: 2510.15780 by Alireza Moradi, Mathieu Tanneau, Pascal Van Hentenryck, Reza Zandehshahvar.

Figure 1
Figure 1. Figure 1: Illustration of the proposed CACP framework. Conformity scores from the calibration set are weighted based on their similarity to the target prediction point—illustrated in purple and orange. While CQR assigns uniform weights to all samples, CACP emphasizes more similar instances, resulting in tighter and more efficient prediction intervals. III. CONTEXT-AWARE CONFORMAL PREDICTION CACP is proposed as a fam… view at source ↗
Figure 2
Figure 2. Figure 2: Coverage vs. AIW trade-off for different CP methods [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distribution of conformity scores for the MISO system￾level forecasts using CQR and CACP at two representative time steps (12pm and 6pm) on 2019-04-2. While the CQR distribution remains relatively stable across hours, the CACP distribution varies with time, showing a right-skewed (more positive) pattern in the evening (6pm) similar to CQR, and a left-shifted (more negative) distribution at noon (12pm). REF… view at source ↗
Figure 3
Figure 3. Figure 3: Conditional coverage for different hours of the day [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

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

40 extracted references · 1 linked inside Pith

  1. [1]

    Conformal prediction for time series with modern hopfield networks.Advances in Neural Informa- tion Processing Systems, 36:56027–56074, 2023

    Andreas Auer, Martin Gauch, Daniel Klotz, and Sepp Hochreiter. Conformal prediction for time series with modern hopfield networks.Advances in Neural Informa- tion Processing Systems, 36:56027–56074, 2023

    2023

  2. [2]

    Conformal prediction beyond exchangeability.The Annals of Statistics, 51(2):816–845, 2023

    Rina Foygel Barber, Emmanuel J Candes, Aaditya Ram- das, and Ryan J Tibshirani. Conformal prediction beyond exchangeability.The Annals of Statistics, 51(2):816–845, 2023

    2023

  3. [3]

    Confor- malized time series with semantic features.Advances in Neural Information Processing Systems, 37:121449– 121474, 2024

    Baiting Chen, Zhimei Ren, and Lu Cheng. Confor- malized time series with semantic features.Advances in Neural Information Processing Systems, 37:121449– 121474, 2024

    2024

  4. [4]

    Data- driven scenario generation of renewable energy produc- tion based on controllable generative adversarial net- works with interpretability.Applied Energy, 308:118387, 2022

    Wei Dong, Xianqing Chen, and Qiang Yang. Data- driven scenario generation of renewable energy produc- tion based on controllable generative adversarial net- works with interpretability.Applied Energy, 308:118387, 2022

    2022

  5. [5]

    Probabilistic solar power forecasting using bayesian model averaging.IEEE Transactions on Sustainable Energy, 12(1):325–337, 2020

    Kate Doubleday, Stephen Jascourt, William Kleiber, and Bri-Mathias Hodge. Probabilistic solar power forecasting using bayesian model averaging.IEEE Transactions on Sustainable Energy, 12(1):325–337, 2020

    2020

  6. [6]

    Adaptive conformal inference under distribution shift.Advances in Neural Information Processing Systems, 34:1660–1672, 2021

    Isaac Gibbs and Emmanuel Candes. Adaptive conformal inference under distribution shift.Advances in Neural Information Processing Systems, 34:1660–1672, 2021

    2021

  7. [7]

    Probabilistic forecasts, calibration and sharp- ness.Journal of the Royal Statistical Society Series B: Statistical Methodology, 69(2):243–268, 2007

    Tilmann Gneiting, Fadoua Balabdaoui, and Adrian E Raftery. Probabilistic forecasts, calibration and sharp- ness.Journal of the Royal Statistical Society Series B: Statistical Methodology, 69(2):243–268, 2007

    2007

  8. [8]

    Probabilistic forecasting.Annual Review of Statistics and Its Applica- tion, 1(1):125–151, 2014

    Tilmann Gneiting and Matthias Katzfuss. Probabilistic forecasting.Annual Review of Statistics and Its Applica- tion, 1(1):125–151, 2014

    2014

  9. [9]

    Probabilistic solar forecasting: Benchmarks, 9 post-processing, verification.Solar Energy, 252:72–80, 2023

    Tilmann Gneiting, Sebastian Lerch, and Benedikt Schulz. Probabilistic solar forecasting: Benchmarks, 9 post-processing, verification.Solar Energy, 252:72–80, 2023

    2023

  10. [10]

    Very short-term nonparametric probabilistic forecasting of renewable energy generation—with application to solar energy.IEEE Transactions on Power Systems, 31(5):3850–3863, 2016

    Faranak Golestaneh, Pierre Pinson, and Hoay Beng Gooi. Very short-term nonparametric probabilistic forecasting of renewable energy generation—with application to solar energy.IEEE Transactions on Power Systems, 31(5):3850–3863, 2016

    2016

  11. [11]

    Probabilistic en- ergy forecasting: Global energy forecasting competition 2014 and beyond, 2016

    Tao Hong, Pierre Pinson, Shu Fan, Hamidreza Zareipour, Alberto Troccoli, and Rob J Hyndman. Probabilistic en- ergy forecasting: Global energy forecasting competition 2014 and beyond, 2016

    2014

  12. [12]

    Energy forecasting: A review and outlook.IEEE Open Access Journal of Power and Energy, 7:376–388, 2020

    Tao Hong, Pierre Pinson, Yi Wang, Rafał Weron, Dazhi Yang, and Hamidreza Zareipour. Energy forecasting: A review and outlook.IEEE Open Access Journal of Power and Energy, 7:376–388, 2020

    2020

  13. [13]

    A novel day-ahead regional and probabilistic wind power forecasting framework using deep cnns and conformalized regression forests

    Jef Jonkers, Diego Nieves Avendano, Glenn Van Wallen- dael, and Sofie Van Hoecke. A novel day-ahead regional and probabilistic wind power forecasting framework using deep cnns and conformalized regression forests. Applied Energy, 361:122900, 2024

    2024

  14. [14]

    Kernel-based optimally weighted conformal prediction intervals.arXiv preprint arXiv:2405.16828, 2024

    Jonghyeok Lee, Chen Xu, and Yao Xie. Kernel-based optimally weighted conformal prediction intervals.arXiv preprint arXiv:2405.16828, 2024

    arXiv 2024

  15. [15]

    Windgmmn: Scenario forecasting for wind power using generative moment matching networks.IEEE transac- tions on artificial intelligence, 3(5):843–850, 2021

    Wenlong Liao, Zhe Yang, Xinxin Chen, and Yaqi Li. Windgmmn: Scenario forecasting for wind power using generative moment matching networks.IEEE transac- tions on artificial intelligence, 3(5):843–850, 2021

    2021

  16. [16]

    Use of physics to improve solar forecast: Part ii, machine learning and model interpretability.Solar Energy, 244:362–378, 2022

    Weijia Liu, Yangang Liu, Tao Zhang, Yongxiang Han, Xin Zhou, Yu Xie, and Shinjae Yoo. Use of physics to improve solar forecast: Part ii, machine learning and model interpretability.Solar Energy, 244:362–378, 2022

    2022

  17. [17]

    Assessing the perfor- mance of deep learning models for multivariate proba- bilistic energy forecasting.Applied Energy, 285:116405, 2021

    Aleksei Mashlakov, Toni Kuronen, Lasse Lensu, Arto Kaarna, and Samuli Honkapuro. Assessing the perfor- mance of deep learning models for multivariate proba- bilistic energy forecasting.Applied Energy, 285:116405, 2021

    2021

  18. [18]

    Enhanced renewable energy forecasting and operations through probabilistic forecast aggregation.arXiv preprint arXiv:2502.07010, 2025

    Alireza Moradi, Mathieu Tanneau, Reza Zandehshahvar, and Pascal Van Hentenryck. Enhanced renewable energy forecasting and operations through probabilistic forecast aggregation.arXiv preprint arXiv:2502.07010, 2025

    Pith/arXiv arXiv 2025

  19. [19]

    Renewable energy sources—modeling and forecasting

    Juan M Morales, Antonio J Conejo, Henrik Madsen, Pierre Pinson, and Marco Zugno. Renewable energy sources—modeling and forecasting. InIntegrating Re- newables in Electricity Markets: Operational Problems, pages 15–56. Springer, 2013

    2013

  20. [20]

    Wind power forecasting using attention-based gated recurrent unit network.Energy, 196:117081, 2020

    Zhewen Niu, Zeyuan Yu, Wenhu Tang, Qinghua Wu, and Marek Reformat. Wind power forecasting using attention-based gated recurrent unit network.Energy, 196:117081, 2020

    2020

  21. [21]

    En- hancing the reliability of probabilistic pv power forecasts using conformal prediction.Solar Energy Advances, 4:100059, 2024

    Yvet Renkema, Lennard Visser, and Tarek AlSkaif. En- hancing the reliability of probabilistic pv power forecasts using conformal prediction.Solar Energy Advances, 4:100059, 2024

    2024

  22. [22]

    Conformalized quantile regression.Advances in neural information processing systems, 32, 2019

    Yaniv Romano, Evan Patterson, and Emmanuel Candes. Conformalized quantile regression.Advances in neural information processing systems, 32, 2019

    2019

  23. [23]

    ARPA-E PERFORM datasets, 2022

    Brian Sergi et al. ARPA-E PERFORM datasets, 2022

    2022

  24. [24]

    Direct interval forecast of uncertain wind power based on recur- rent neural networks.IEEE Transactions on Sustainable Energy, 9(3):1177–1187, 2017

    Zhichao Shi, Hao Liang, and Venkata Dinavahi. Direct interval forecast of uncertain wind power based on recur- rent neural networks.IEEE Transactions on Sustainable Energy, 9(3):1177–1187, 2017

    2017

  25. [25]

    Soman, Hamidreza Zareipour, Om Malik, and Paras Mandal

    Saurabh S. Soman, Hamidreza Zareipour, Om Malik, and Paras Mandal. A review of wind power and wind speed forecasting methods with different time horizons. InNorth American Power Symposium 2010, pages 1–8, 2010

    2010

  26. [26]

    A review of wind power and wind speed forecasting methods with different time horizons

    Saurabh S Soman, Hamidreza Zareipour, Om Malik, and Paras Mandal. A review of wind power and wind speed forecasting methods with different time horizons. InNorth American power symposium 2010, pages 1–8. IEEE, 2010

    2010

  27. [27]

    The future of forecasting for renewable energy.Wiley Interdisciplinary Reviews: Energy and Environment, 9(2):e365, 2020

    Conor Sweeney, Ricardo J Bessa, Jethro Browell, and Pierre Pinson. The future of forecasting for renewable energy.Wiley Interdisciplinary Reviews: Energy and Environment, 9(2):e365, 2020

    2020

  28. [28]

    A review of very short-term wind and solar power forecasting.Renewable and Sustainable Energy Reviews, 153:111758, 2022

    Rosemary Tawn and Jethro Browell. A review of very short-term wind and solar power forecasting.Renewable and Sustainable Energy Reviews, 153:111758, 2022

    2022

  29. [29]

    Conformal prediction under covariate shift.Advances in neural information process- ing systems, 32, 2019

    Ryan J Tibshirani, Rina Foygel Barber, Emmanuel Can- des, and Aaditya Ramdas. Conformal prediction under covariate shift.Advances in neural information process- ing systems, 32, 2019

    2019

  30. [30]

    Springer, 2005

    Vladimir V ovk, Alexander Gammerman, and Glenn Shafer.Algorithmic learning in a random world, vol- ume 29. Springer, 2005

    2005

  31. [31]

    Conformal asymmetric multi-quantile generative transformer for day-ahead wind power interval prediction.Applied Energy, 333:120634, 2023

    Wei Wang, Bin Feng, Gang Huang, Chuangxin Guo, Wenlong Liao, and Zhe Chen. Conformal asymmetric multi-quantile generative transformer for day-ahead wind power interval prediction.Applied Energy, 333:120634, 2023

    2023

  32. [32]

    A review of wind speed and wind power forecasting with deep neural networks.Applied energy, 304:117766, 2021

    Yun Wang, Runmin Zou, Fang Liu, Lingjun Zhang, and Qianyi Liu. A review of wind speed and wind power forecasting with deep neural networks.Applied energy, 304:117766, 2021

    2021

  33. [33]

    A stacked gru-rnn-based approach for predict- ing renewable energy and electricity load for smart grid operation.IEEE Transactions on Industrial Informatics, 17(10):7050–7059, 2021

    Min Xia, Haidong Shao, Xiandong Ma, and Clarence W De Silva. A stacked gru-rnn-based approach for predict- ing renewable energy and electricity load for smart grid operation.IEEE Transactions on Industrial Informatics, 17(10):7050–7059, 2021

    2021

  34. [34]

    An overview of deterministic and probabilistic forecasting methods of wind energy

    Yuying Xie, Chaoshun Li, Mengying Li, Fangjie Liu, and Meruyert Taukenova. An overview of deterministic and probabilistic forecasting methods of wind energy. Iscience, 26(1), 2023

    2023

  35. [35]

    Conformal prediction interval for dynamic time-series

    Chen Xu and Yao Xie. Conformal prediction interval for dynamic time-series. InInternational Conference on Machine Learning, pages 11559–11569. PMLR, 2021

    2021

  36. [36]

    Conformal prediction for time se- ries.IEEE transactions on pattern analysis and machine intelligence, 45(10):11575–11587, 2023

    Chen Xu and Yao Xie. Conformal prediction for time se- ries.IEEE transactions on pattern analysis and machine intelligence, 45(10):11575–11587, 2023

    2023

  37. [37]

    Sequential predictive conformal inference for time series

    Chen Xu and Yao Xie. Sequential predictive conformal inference for time series. InInternational Conference on Machine Learning, pages 38707–38727. PMLR, 2023

    2023

  38. [38]

    A short- term wind power forecasting approach with adjustment of numerical weather prediction input by data mining

    Qianyao Xu, Dawei He, Ning Zhang, Chongqing Kang, Qing Xia, Jianhua Bai, and Junhui Huang. A short- term wind power forecasting approach with adjustment of numerical weather prediction input by data mining. 10 IEEE Transactions on sustainable energy, 6(4):1283– 1291, 2015

    2015

  39. [39]

    Weather-informed prob- abilistic forecasting and scenario generation in power systems.Applied Energy, 384:125369, 2025

    Hanyu Zhang, Reza Zandehshahvar, Mathieu Tanneau, and Pascal Van Hentenryck. Weather-informed prob- abilistic forecasting and scenario generation in power systems.Applied Energy, 384:125369, 2025

    2025

  40. [40]

    Review on probabilistic forecasting of wind power generation

    Yao Zhang, Jianxue Wang, and Xifan Wang. Review on probabilistic forecasting of wind power generation. Renewable and Sustainable Energy Reviews, 32:255– 270, 2014

    2014