Dynamic graph structure and spatio-temporal representations in wind power forecasting DOI Creative Commons

Peng Zang,

Wenqi Dong, Jing Wang

et al.

Science and Technology for Energy Transition, Journal Year: 2024, Volume and Issue: 80, P. 9 - 9

Published: Nov. 13, 2024

Wind Power Forecasting (WPF) has gained considerable focus as a crucial aspect of the successful integration and operation wind power. However, due to stochastic unstable nature wind, it poses real challenge effectively analyze correlations among multiple time series data for accurate prediction. In our study, an end-to-end framework called Dynamic Graph structure Spatio-Temporal representation learning (DSTG) is proposed achieve stable power forecasting by constructing graph capture critical features in data. Specifically, Structure Learning (GSL) module introduced dynamically construct task-related correlation matrices via backpropagation mitigate inherent inconsistency randomness Additionally, dual-scale temporal (DTG) further explore implicit spatio-temporal at fine-grained level using different skip connections from constructed Finally, comprehensive experiments are performed on collected Xuji Group (XGWP) dataset, results show that DSTG outperforms state-of-the-art methods 10.12% average root mean square error absolute error, demonstrating effectiveness DSTG. conclusion, model provides promising approach.

Language: Английский

A reconstruction-based secondary decomposition-ensemble framework for wind power forecasting DOI

Runkun Cheng,

Di Yang,

Da Liu

et al.

Energy, Journal Year: 2024, Volume and Issue: 308, P. 132895 - 132895

Published: Aug. 19, 2024

Language: Английский

Citations

7
Short-term subway passenger flow forecasting approach based on multi-source data fusion DOI
Yifan Cheng, Hongtao Li, Shaolong Sun

et al.

Information Sciences, Journal Year: 2024, Volume and Issue: 679, P. 121109 - 121109

Published: Sept. 1, 2024

Language: Английский

Citations

4
Medium-term wind speed ensemble forecasting using ICEEMDAN and dual attention mechanism DOI
Jinhua Liu,

Zhiyuan Leng,

Lu Chen

et al.

Published: Jan. 9, 2025

Wind power output has strong randomness, volatility, and intermittency. To maintain the safety stability of large grid under new system, high-precision medium-term wind forecasting is urgently needed. This paper fully leverages temporal dynamics dataset proposes a ensemble model that integrates transfer entropy, improved complete empirical mode decomposition with adaptive noise (ICEEMDAN) decomposition, dual attention mechanism, multiple recurrent neural networks. Firstly, we compare entropy meteorological factors to determine direction information flow select set characteristic variables. Next, utilizing ICEEMDAN signal algorithm, sequence segmented into various intrinsic functions, attention-based LSTM, GRU, BiLSTM models are established. After aggregation reconstruction, three sets predictions obtained. Finally, mechanism combined dynamically weight achieve predictions. Actual examples show compared several benchmark models, proposed notably enhances predictive accuracy forecasting.

Language: Английский

Citations

0
A Wind and Solar Power Prediction Method Based on Temporal Convolutional Network–Attention–Long Short-Term Memory Transfer Learning and Sensitive Meteorological Features DOI Creative Commons
Yuan Wang, Yue Bi, Guo Yu

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(3), P. 1636 - 1636

Published: Feb. 6, 2025

To address the issue of declining prediction accuracy caused by lack data in newly constructed wind and solar power stations, this paper introduces a transfer learning-based forecasting approach for photovoltaic power. The method incorporates sensitive meteorological feature selection utilizes Temporal Convolutional Network–Attention–Long Short-Term Memory (TCN-ATT-LSTM) model. Spearman’s rank correlation, mutual information entropy, Pearson correlation are employed to investigate relationship between features output. Through evidence theory, with cumulative contribution exceeding 85% selected as inputs TCN-ATT-LSTM network is pre-trained extract common knowledge, learning applied fine-tune (FT) model through parameter adjustments. This enables adaptive be quickly target stations limited data, improving accuracy. Finally, effectiveness proposed validated its application from projected station planned region northwestern China. not only enhances emerging but also has significant implications renewable energy industry.

Language: Английский

Citations

0
A Novel Hybrid Deep Learning Model for Day-Ahead Wind Power Interval Forecasting DOI Open Access

Jianjing Mao,

Jian Zhao, H. Zhang

et al.

Sustainability, Journal Year: 2025, Volume and Issue: 17(7), P. 3239 - 3239

Published: April 5, 2025

Accurate interval forecasting of wind power is crucial for ensuring the safe, stable, and cost-effective operation grids. In this paper, we propose a hybrid deep learning model day-ahead forecasting. The begins by utilizing Gaussian mixture (GMM) to cluster daily data with similar distribution patterns. To optimize input features, feature selection (FS) method applied remove irrelevant data. empirical wavelet transform (EWT) then employed decompose both numerical weather prediction (NWP) into frequency components, effectively isolating high-frequency components that capture inherent randomness volatility A convolutional neural network (CNN) used extract spatial correlations meteorological while bidirectional gated recurrent unit (BiGRU) captures temporal dependencies within sequence. further enhance accuracy, multi-head self-attention mechanism (MHSAM) incorporated assign greater weight most influential elements. This leads development based on GMM-FS-EWT-CNN-BiGRU-MHSAM. proposed validated through comparison benchmark demonstrates superior performance. Furthermore, forecasts generated using NPKDE shows new achieves higher accuracy.

Language: Английский

Citations

0
Physics-informed reinforcement learning for probabilistic wind power forecasting under extreme events DOI
Yanli Liu, Junyi Wang,

L. Liu

et al.

Applied Energy, Journal Year: 2024, Volume and Issue: 376, P. 124068 - 124068

Published: Aug. 23, 2024

Language: Английский

Citations

3
Wind power forecasting using optimized LSTM by attraction–repulsion optimization algorithm DOI Creative Commons
Mohammed A. A. Al‐qaness, Ahmed A. Ewees, Ahmad O. Aseeri

et al.

Ain Shams Engineering Journal, Journal Year: 2024, Volume and Issue: unknown, P. 103150 - 103150

Published: Nov. 1, 2024

Language: Английский

Citations

3
Fully connected multi-reservoir echo state networks for wind power prediction DOI
Xianshuang Yao,

Kangshuai Guo,

Jianlin Lei

et al.

Energy, Journal Year: 2024, Volume and Issue: unknown, P. 133579 - 133579

Published: Oct. 1, 2024

Language: Английский

Citations

2
Proactive failure warning for wind power forecast models based on volatility indicators analysis DOI
Yunxiao Chen,

Chaojing Lin,

Yilan Zhang

et al.

Energy, Journal Year: 2024, Volume and Issue: 305, P. 132310 - 132310

Published: July 3, 2024

Language: Английский

Citations

2
Combined Prediction of Wind Power in Extreme Weather Based on Time Series Adversarial Generation Networks DOI Creative Commons
Wenjie Ye, Dongmei Yang,

Chenghong Tang

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 102660 - 102669

Published: Jan. 1, 2024

The current global climate is complex with an increasing frequency of extreme weather events. randomness, variability, and intermittency new energy sources pose significant challenges for the balance between power generation consumption in electricity grids. This challenge particularly pronounced during events such as cold waves, which aggravate supply stability. Accurate prediction wind output can reduce need system backup capacity, ensuring stable operation reliability system. However, models do not effectively consider impact weather, leading to low accuracy large deviations. Additionally, mechanisms by affects differ from those under normal conditions. Extreme waves are rare, sample data scarce, making it difficult establish precise models. To address these issues, this study proposes a correction method based on time-series generative adversarial networks. First, network algorithm was used generate samples meteorological data, constructing database issue scarcity. Second, improved particle swarm (IPSO) Bayesian optimization (BO) were optimize parameters single-prediction algorithms eXtreme gradient boosting (XGBoost) least absolute shrinkage selection operator (LASSO). Subsequently, combining performance principle maximum diversity, optimal combination determined using evaluation indicators error Pearson correlation coefficients construct model stacking ensemble learning framework, overcoming limitations single improving accuracy. Finally, Support Vector Regression (SVR) similar days, correcting predictions. Through verification actual wave case study, proposed demonstrated significantly compared conventional

Language: Английский

Citations

2