Pith. sign in

REVIEW 5 major objections 6 minor 26 references

Applying Ensemble Models based on Graph Neural Network and Reinforcement Learning for Wind Power Forecasting

T0 review · 5 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read An ensemble of GNN and RL cuts wind forecast error by up to 12.89%

desk verdict The ensemble idea is sensible, but the central tables are internally inconsistent, so the empirical claim cannot be evaluated as printed. read the letter →

arxiv 2501.16591 v1 pith:EEUUK4N4 submitted 2025-01-28 cs.LG cs.AI

Hongjin Song , Qianrun Chen , Tianqi Jiang , Yongfeng Li , Xusheng Li , Wenjun Xi , Songtao Huang This is my paper · ORCID
classification cs.LGcs.AI
keywords windpowerforecastingensemblelearninggraphneuralnetworkreinforcementactor-criticstateembeddingNRELdatasetGEFC
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 tries to show that wind power forecasts improve when no single model is trusted: a graph neural network embeds the target wind farm's history together with neighboring farms, and an actor-critic reinforcement learning agent reweights four base models (ARIMA, LightGBM, LSTM, and SGNN) at every time step. If true, this means the question 'which model is best right now?' is as important as the question 'what is the next power value?', and both can be answered by one learned ensemble. The reported result is up to a 12.89% lower error than the best baseline on the NREL and GEFC open wind datasets.

What carries the argument

The state embedding is the load-bearing piece. It is built in two parallel branches: a spatio-temporal branch passes each wind farm's dilated-CNN-compressed time series through a GNN whose message-passing stages (computation, aggregation, update) let a target farm's representation absorb information from geographically neighboring farms; a model-loss branch compresses the recent prediction losses of all base models on the target farm through a multi-layer perceptron. The two vectors are concatenated into $SE_A = (STSE_A, MLE_A)$, and this state is fed into an actor-critic loop where the actor outputs weights $\mathbf{a}_t = [W_{t,1}, \dots, W_{t,N}]$ and the reward is derived from comparing the ensemble forecast with the observed power.

What would settle it

Train the same EMGRL on a wind farm group where the geographically closest farms are strongly anti-correlated in power output (for example, a ridge with different wind exposures), and compare test error against an EMGRL variant with the GNN removed. If the GNN variant does not beat the no-GNN variant, the geographic-graph assumption fails. A simpler observational check: compute the sample cross-correlation of wind power between nearest-neighbor farms in the GEFC dataset; if most neighbor pairs show near-zero or negative correlation, message passing across the geographic graph cannot be the source of the reported gain.

Watch

Extended reading notes

Core claim

The central claim is that an ensemble whose weights are chosen by reinforcement learning, and whose state representation is a spatio-temporal embedding of wind farms produced by a GNN, forecasts wind power more accurately than any single time-series model and than fixed-weight ensembles. The paper constructs a graph whose nodes are wind farms and whose edges encode geographical proximity; a dilated CNN compresses each farm's recent power sequence into a vector, the GNN mixes these vectors across neighboring farms, and the resulting embedding is concatenated with a compressed vector of the base models' recent losses. This state is fed to an actor-critic agent that outputs normalized weights for the four base models, and the weighted sum becomes the forecast. On four NREL offshore wind farms the method reports RMSE reductions of roughly 6–10% over the best single base model, and on seven GEFC wind farms MAE around 7.74% lower than the best baseline SDAE, with the overall best case reaching 12.89% improvement.

Load-bearing premise

The method assumes that the right way to use other wind farms is to let geographically nearest farms speak to each other through the GNN; if distance is not a reliable proxy for the wind-power dependence that carries forecasting information, the entire state embedding loses its advantage.

Editorial extensions

If this is right

  • If EMGRL is right, forecasters do not need to build one perfect model; they can maintain a portfolio of diverse models and learn when each one leads.
  • The GNN's neighborhood aggregation makes the forecast for one farm depend explicitly on other farms' recent output, so the method's accuracy should scale with the density of the wind farm network.
  • Because weights are re-derived at every timestep from recent losses, the ensemble can track non-stationary wind conditions and data shifts better than fixed-weight stacking.
  • The 12.89% best-case improvement, if reproducible, would translate into reduced reserve requirements and more reliable grid scheduling for wind operators.

Reading between the lines

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

  • The graph edge weights are based purely on geographic distance, but wind power dependence actually propagates with weather systems; a natural testable extension is to orient or weight edges by prevailing wind direction or by lagged cross-correlation between farms.
  • The same actor-critic-over-base-models recipe could be transferred to other spatiotemporal forecasting domains (solar power, load forecasting, traffic) wherever base models shift in relative performance.
  • The 12.89% figure comes from a specific best-farm comparison; a more demanding check would be to average the improvement over all farms and test statistical significance, which the paper does not report.
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

5 major / 6 minor

Summary. The paper proposes EMGRL, an ensemble model for wind power forecasting that combines ARIMA, LightGBM, LSTM, and SGNN as base learners. A dilated CNN compresses each farm's time series, a GNN aggregates neighboring-farm representations, and an actor-critic reinforcement learning module outputs ensemble weights from a state embedding that includes the historical losses of the base models. The experiments on the NREL and GEFC 2012 datasets claim that EMGRL outperforms seven baselines by up to 12.89% in MAE and RMSE. The core evaluation section, however, contains internal inconsistencies that undermine the reported quantitative claims.

Significance. The idea of using GNN-based spatial state embeddings and actor-critic weight selection is a reasonable and potentially useful direction for wind power forecasting ensembles, and the use of two open datasets is appropriate. The paper also separates training and test years and fits the ensemble weights on training data, so the headline result is not circular. Nevertheless, the empirical validation as presented cannot be trusted: the RMSE table is identical to the MAE table, which is impossible for nonzero forecast errors, and the claimed improvement percentages do not match the numbers shown. The contribution will be significant only after the experiments are redone or correctly reported; as submitted, the central claim is unsupported.

major comments (5)
  1. [§5.5, Tables 2 and 3] Table 2 (labeled MAE) and Table 3 (labeled RMSE) on the GEFC dataset are byte-for-byte identical for every model and every wind farm. For any nontrivial forecast error, RMSE must be strictly greater than MAE, so at least one of these tables cannot be reporting the metric named in its caption. This invalidates the GEFC numerical evidence and any conclusion drawn from it.
  2. [§5.5, Table 1 and Figures 10–11] The metric for the NREL experiment is ambiguous: Table 1's caption says MAE, the text says RMSE, Figure 10 says MAE, and Figure 11 says RMSE. In addition, the improvements over SGNN quoted in the text (9.62%, 10.44%, 6.57%, 6.34% for farms A–D) do not match the values computed from Table 1 (8.78%, 9.45%, 6.17%, 5.97%). The reported improvements are therefore not reproducible from the numbers as printed.
  3. [Abstract and §6] The headline claim of an improvement of up to 12.89% is not derivable from any reported comparison. The largest improvement in Table 1 is about 9.45% (if MAE is intended), and the GEFC improvements over the best baseline are roughly 4.9–7.7% depending on the farm. No table, equation, or calculation in the paper yields 12.89%, so the central quantitative claim lacks support.
  4. [§5.4–5.5] No error bars, standard deviations, numbers of independent runs, or statistical significance tests are reported for any result. Because the claimed margins over SGNN and SDAE are only a few percent, these margins could easily fall within run-to-run variability, so the central comparison is not statistically supported as presented.
  5. [§4.4, Algorithm 1] The training algorithm is not fully specified: line 3 samples a'_t = π_θ(s,a) with an undefined action a, line 5 writes the actor update as θ = θ + α log π_θ(s_t,a_t) Q_ω(s_t,a_t), omitting the gradient operator and any baseline, and no hyperparameters, network sizes, or optimizer settings are given in Section 5.3. This makes the method irreproducible as described.
minor comments (6)
  1. [§4.2, Eq. (7)] The text says the base-model losses are compressed by a dilated convolutional neural network, but Eq. (7) uses an MLP; the notation and description should be aligned.
  2. [§4.2, Eq. (6)] The notation t_k in Eq. (6) is not defined; the input window and features fed to the dilated CNN should be specified.
  3. [§4.4, Algorithm 1] The symbol a_t^2 and the line numbering in Algorithm 1 are garbled; the action should be written consistently, for example as a'_t.
  4. [§1 and §4.2] The introduction says all turbines within a wind farm are graph nodes, but Section 4.2 builds a graph over wind farms A–F; the node semantics should be made consistent.
  5. [References] Several references have formatting problems, including mixed punctuation and incomplete author lists; the reference list should be checked against the journal style.
  6. [§5] No code or data availability statement is included, and no explicit hyperparameter settings are provided, which is essential for reproducing the experimental tables.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: EMGRL's ensemble weights are trained on disjoint data and no load-bearing self-citation or definitional reduction is present.

full rationale

The paper's central claim is empirical: EMGRL outperforms baseline wind-power forecasting models on the NREL and GEFC datasets. The derivation chain is a standard train/test pipeline: base models are trained on the training period, the RL actor-critic assigns ensemble weights using a state embedding that combines GNN-processed spatiotemporal features and historical base-model losses, and test errors are reported on held-out periods (NREL 2011; GEFC January 2011 through June 2012). The inclusion of historical base-model loss in the state is a design choice, not a definitional reduction: the target quantity (future test error on held-out data) is not used to construct the embedding or the weights. No load-bearing self-citation appears: SGNN is cited as a baseline, not as a justification for the method, and the paper invokes no uniqueness theorem, no ansatz smuggled in via the authors' prior work, and no renaming of a known result as a new derivation. The serious numerical inconsistency between Tables 2 and 3 (the RMSE table is byte-for-byte identical to the MAE table, which is impossible for real forecasts because RMSE >= MAE) is an evidence-quality or correctness defect, not circularity: the claim does not reduce to its own inputs by construction. Since no specific circular step can be quoted and exhibited, the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities; its novel elements are model components (state embedding, RL weighting), not ontological claims. The free-parameter ledger is dominated by unreported architecture and training choices.

free parameters (5)
  • Base model hyperparameters (ARIMA order, LightGBM structure, LSTM hidden size, SGNN depth)
    Each base model is trained on historical wind data, but exact configurations are not reported, so the ensemble depends on unspecified tuned values.
  • Dilated CNN architecture and kernel settings
    Used to compress time-series and loss sequences into embeddings (Section 4.2); no layer counts or dilation rates are given.
  • GNN architecture (message passing layers, hidden dimensions)
    Aggregates neighboring wind farm embeddings (Section 4.2, Eq. 6); details are absent.
  • Actor-critic network architecture and learning rates
    The policy and value networks that assign ensemble weights are not specified (Section 4.3).
  • RL hyperparameters (discount factor, replay buffer size, batch size)
    Required by Algorithm 1 but never given.
assumptions (4)
  • domain assumption Wind farms connected by geographic proximity share predictive information about wind power (Section 4.2, Fig. 6).
    The GNN aggregates neighbor embeddings based on location; if distance is not the right dependency structure, the state embedding is misleading.
  • domain assumption The historical loss profile of each base model is informative for choosing future ensemble weights (Section 4.2, Eq. 7).
    The model-loss embedding assumes yesterday's relative performance predicts today's best model.
  • domain assumption The state embedding satisfies the Markov property for the RL problem (Section 4.3).
    Actor-critic MDP requires that the embedding of current time series plus loss history is a sufficient statistic for optimal weight selection.
  • domain assumption The temporal train/test splits (2010 vs 2011 for NREL, 2009-2010 vs 2011-2012 for GEFC) have no distribution shift that invalidates transfer (Section 5.1).
    If wind patterns shift between years, the learned weights may not transfer; no drift analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Applying Ensemble Models based on Graph Neural Network and Reinforcement Learning for Wind Power Forecasting." pith.science (2026). https://pith.science/paper/EEUUK4N4

@misc{pith2026250116591,
  author       = {Pith},
  title        = {Pith review of: Applying Ensemble Models based on Graph Neural Network and Reinforcement Learning for Wind Power Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EEUUK4N4}},
  note         = {Machine review of arXiv:2501.16591}
}
read the original abstract

Accurately predicting the wind power output of a wind farm across various time scales utilizing Wind Power Forecasting (WPF) is a critical issue in wind power trading and utilization. The WPF problem remains unresolved due to numerous influencing variables, such as wind speed, temperature, latitude, and longitude. Furthermore, achieving high prediction accuracy is crucial for maintaining electric grid stability and ensuring supply security. In this paper, we model all wind turbines within a wind farm as graph nodes in a graph built by their geographical locations. Accordingly, we propose an ensemble model based on graph neural networks and reinforcement learning (EMGRL) for WPF. Our approach includes: (1) applying graph neural networks to capture the time-series data from neighboring wind farms relevant to the target wind farm; (2) establishing a general state embedding that integrates the target wind farm's data with the historical performance of base models on the target wind farm; (3) ensembling and leveraging the advantages of all base models through an actor-critic reinforcement learning framework for WPF.

Figures

Figures reproduced from arXiv: 2501.16591 by the authors.

Figure 1
Figure 1. The overall framework of ensemble learning. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. The Message aggregation process in GNN for node [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Markov Decision Process (MDP) The parameter space of an MDP can be represented as a tuple (𝑆, 𝐴, 𝑃, 𝑅, γ), where: 𝑆 is a finite set of states with 𝑠! ∈ 𝑆, where 𝑠! is the state at time 𝑡; 𝐴 is a finite set of actions, with 𝑎! ∈ 𝐴, where 𝑎! is a specific action; 𝑃 represents the state transition probabilities, 𝑃+( 𝑠!"* ∣ 𝑠!, 𝑎! ), predicting the next state 𝑠!"* based on the current state 𝑠! and action 𝑎!. 𝑃+ indicate… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Ensemble learning model decision process. The main objective of wind power prediction is to forecast wind power over a period based on historical wind power data from a specific area (wind farm) and the wind power characteristics of neighboring areas. For time series d…
Figure 5
Figure 5. Figure 5: The overall framework of EMGRL model. The overall framework of the EMGRL model is shown in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: A flowchart for generating state embedding vectors [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Reinforcement learning model training process [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Abstract map of wind farm location in NREL Dataset. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Abstract map of wind farm location in GEFC [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: MAE performance comparison between EMGRL and multiple baseline methods on NREL [PITH_FULL_IMAGE:figures/full_fig_p019_10.png]
Figure 11
Figure 11. Figure 11: Improvement of EMGRL model on RMSE compared to the best [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: MAE performance comparison between EMGRL and multiple baseline methods on GEFC [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]
Figure 13
Figure 13. Figure 13: Improvement of EMGRL model on MAE compared to the best [PITH_FULL_IMAGE:figures/full_fig_p021_13.png]
Figure 14
Figure 14. Figure 14: RMSE performance comparison between EMGRL and multiple baseline methods on GEFC [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]
Figure 15
Figure 15. Figure 15: Improvement of EMGRL model on RMSE compared to the best [PITH_FULL_IMAGE:figures/full_fig_p022_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 25 canonical work pages

  1. [1]

    The WPF problem remains unresolved due to numerous influencing variables, such as wind speed, temperature, latitude, and longitude

    Applying Ensemble Models based on Graph Neural Network and Reinforcement Learning for Wind Power Forecasting Hongjin Songa , Qianrun Chenb, Tianqi Jiangc, Yongfeng Lid, Xusheng Lie, Wenjun Xif and Songtao Huangb,* aNorthwestern University bLanzhou University cSouth China University of Technology dXi’an Jiaotong University eCloud Gansu Technology Co., Ltd ...

    work page 2022

  2. [5]

    Firstly, we propose a method for generating state embeddings that integrates temporal data and spatial features of wind farms using a dilated convolutional neural network and a graph neural network, further combined with model loss to generate embedding vectors. Secondly, we input the obtained embedding vector into an actor-critic reinforcement learning m...

    work page 2015

  3. [6]

    #!=𝑃! (1) where the wind power 𝑃!

    2 Related Works 2.1 Traditional Methods Compared to machine learning methods, traditional methods emphasize wind power prediction through rule-based modeling. These traditional approaches can be classified into persistence, physical, and statistical methods. 2.1.1 Persistence Methods The general idea of persistence forecasting is to use the average value ...

    work page 2008

  4. [8]

    The hyperparameters of the SVM are optimized using the JAYA optimization algorithm using the most representative features in the input data

    proposed a support vector machine (JAYA-SVM) short-term wind speed prediction model based on the JAYA algorithm. The hyperparameters of the SVM are optimized using the JAYA optimization algorithm using the most representative features in the input data. In addition to SVM, traditional machine learning algorithms such as KNN (Mahaseth et al., 2022), XGBoos...

    work page 2022

  5. [11]

    Table 1: MAE values of different baseline methods and EMGRL models on NREL Datasets. Model Wind Farm A Wind Farm B Wind Farm C Wind Farm D ARIMA 1.5465 1.4618 1.4594 1.3982 SVR 1.2954 1.2235 1.2300 1.1832 LightGBM 1.1698 1.1214 1.1283 1.0847 LSTM 1.3846 1.2982 1.3034 1.2750 SGNN 1.1520 1.1416 1.1184 1.0722 SDAE 1.3326 1.2143 1.1905 1.1761 COBRT 1.2517 1.1...

    work page 1905

  6. [12]

    The first part involves using real-time wavelet packet decomposition enhanced deep echo state network to construct the basic model with different vanishing moments

    proposed a dynamic integrated wind speed forecasting model based on deep reinforcement learning, considering the time-varying characteristics of wind speed series. The first part involves using real-time wavelet packet decomposition enhanced deep echo state network to construct the basic model with different vanishing moments. The second part determines t...

    work page 2019

  7. [13]

    Ensemble learning methods can generally be subdivided into Bootstrap aggregating (Bagging), Boosting, and Stacking. The bagging method randomly samples training subsets from the training set and then uses these subsets to perform parallel training while integrating the base learners into the ensemble model. The Boosting method first trains a basic learner...

    work page 1998

  8. [15]

    5.1.2 GEFC Dataset The second Dataset we use is from Global Energy Forecasting Competition (GEFC) 2012 - Wind Forecasting

    Figure 8: Abstract map of wind farm location in NREL Dataset. 5.1.2 GEFC Dataset The second Dataset we use is from Global Energy Forecasting Competition (GEFC) 2012 - Wind Forecasting. The GEFC dataset includes three years of historical data from seven wind farms, spanning from the first hour of July 1, 2009 to the 12th hour of June 28,

    work page 2012

Show all 26 references
  1. [18]

    These findings again demonstrate the superiority of our model, which performs well across multiple datasets

    Similarly, on the GEFC dataset (Table 3, Figure 14 and 15), the RMSE of the EMGRL model also achieves optimal results. These findings again demonstrate the superiority of our model, which performs well across multiple datasets. Table 2: MAE values of different baseline methods...

    2008

  2. [20]

    Dayan, P., & Watkins, C

    Global Wind Energy Council: Brussels, Belgium, . Dayan, P., & Watkins, C. (1992). Q-learning. Machine learning, 8, 279–292. Demolli, H., Dokuz, A. S., Ecemis, A., & Gokcek, M. (2019). Wind power forecasting based on daily wind speed data using machine learning algorithms. Ener...

    1992

  3. [23]

    Lahouar, A., & Slama, J. B. H. (2017). Hour-ahead wind power forecast based on random forests. Renewable energy, 109, 529–541. Li, G., & Chiang, H.-D. (2016). Toward cost-oriented forecasting of wind power generation. IEEE Transactions on Smart Grid, 9, 2508–2517. Lillicrap, T...

    2017 arXiv

  4. [24]

    Wang, Y., Hu, Q., Meng, D., & Zhu, P. (2017). Deterministic and probabilistic wind power forecasting using a variational bayesian-based adaptive robust multi-kernel regression model. Applied energy,

    2017

  5. [26]

    Wu, Q., Guan, F., Lv, C., & Huang, Y

    Available online: https://www.who.int/westernpacific/health-topics/air-pollution (accessed on 10 October 2022). Wu, Q., Guan, F., Lv, C., & Huang, Y. (2021). Ultra-short-term multi-step wind power forecasting based on cnn-lstm. IET Renewable Power Generation, 15,. Wu, Q., Zhen...

    2021

  6. [161]

    Chen, C., & Liu, H. (2021). Dynamic ensemble wind speed prediction model based on hybrid deep reinforcement learning. Advanced Engineering Informatics, 48, 101290. Council, G. W. E. (2022). Global wind report

    2021

  7. [208]

    Wang, Y., Zou, R., Liu, F., Zhang, L., & Liu, Q. (2021). A review of wind speed and wind power forecasting with deep neural networks. Applied Energy, 304, 117766. Welch, G. F. (2020). Kalman filter. Computer Vision: A Reference Guide, (pp. 1–3). WHO

    2021

  8. [273]

    S., Maragatham, G., Boopathi, K., & Rangaraj, A

    Devi, A. S., Maragatham, G., Boopathi, K., & Rangaraj, A. (2020). Hourly day-ahead wind power forecasting with the eemd-cso-lstm-efg deep learning technique. Soft Computing, 24, 12391–12411. Dhiman, H. S., Anand, P., & Deb, D. (2019). Wavelet transform and variants of svr with...

    2020

  9. [2010]

    To address the low performance of the NWP prediction algorithm, Wang et al

    After validating with actual electricity data in the Czech Republic, they found that the model could make accurate predictions within 72 hours (Pelikan et al., 2010). To address the low performance of the NWP prediction algorithm, Wang et al. (Wang et al.,

    2010

  10. [2011]

    Although current methods have achieved significant success for WPF, numerous variables (temperature, altitude, position, humidity, pressure, etc.) could influence the results

    and machine learning methods (Demolli et al., 2019; Jørgensen & Shaker, 2020). Although current methods have achieved significant success for WPF, numerous variables (temperature, altitude, position, humidity, pressure, etc.) could influence the results. Each methods has its a...

    2019

  11. [2012]

    In this dataset, the period from July 1, 2009 to December 31, 2010, is used for model training, and the period from January 1, 2011 to June 28, 2012 is used for model testing

    The historical data consists of hourly wind power measurements for each wind farm and forecasts of zonal and meridional wind components, wind speed, and direction for the next 1 to 48 hours. In this dataset, the period from July 1, 2009 to December 31, 2010, is used for model ...

    2009

  12. [2017]

    proposed a generalized multi-core regression model and introduced the variational Bayes method to improve the model. This was done to address the shortcomings of traditional wind power prediction models, which are not sensitive to high-resolution data and have assumptions inco...

    2020

  13. [2018]

    However, traditional machine learning methods require manual extraction and cleaning of data, followed by feature engineering

    have also been used in wind power prediction tasks. However, traditional machine learning methods require manual extraction and cleaning of data, followed by feature engineering. The selection of features largely determines the model's effectiveness. Therefore, traditional mac...

    2019

  14. [2019]

    This algorithm demonstrated excellent generalization ability for a variety of models, proving the physical method's applicability

    proposed an NWP speed correction algorithm based on the Sequence Transfer Correction Algorithm (STCA). This algorithm demonstrated excellent generalization ability for a variety of models, proving the physical method's applicability. The results obtained by the physical method...

    2020

  15. [2020]

    The ARIMA technique is commonly utilized in time series forecasting tasks, including wind power forecasting

    and Bayesian model (Pole et al., 2018), which aim to determine the relationship between input variables and output using historical datasets. The ARIMA technique is commonly utilized in time series forecasting tasks, including wind power forecasting. The usual steps of ARIMA i...

    2018

  16. [2021]

    introduced an attention-based gated recurrent unit sequence model (AGRU) to improve prediction performance, using hidden GRU modules to connect and activate tasks across different prediction steps. A new data-driven model based on the principles of deep learning-convolutional ...

    2021

  17. [2022]

    and Random Forest (RF) (Lahouar & Slama, 2017; Shi et al.,

    2017

  18. [3764]

    , & R i o f l o r i d o , C

    Hong, Y .-Y. , & R i o f l o r i d o , C . L . P. P. ( 2 0 1 9 ) . A h y b r i d d e e p l e a r n i n g-based neural network for 24-h ahead wind power forecasting. Applied Energy, 250, 530–539. Jalali, S. M. J., Os ́orio, G. J., Ahmadian, S., Lotfi, M., Campos, V . M., Shafie...

    2021

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.