A wind power prediction method based on CEEMDAN- LightGBM-Transformer

In the development of green energy, wind power, as the core clean energy, has the inherent intermittency, volatility and randomness of its output, which is very easy to cause the power imbalance of the energy system and the surge of grid accommodation pressure, seriously restricting the safe and stable operation of the energy system. This study proposes the CEEMDAN-LightGBM-Transformer model. First, Complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) decomposes the data into multiple intrinsic mode function(IMF), and uses the "drop_k" strategy to select IMF components with prominent feature information and low redundancy, effectively stripping the noise and invalid fluctuations in wind power output data, and consolidating the foundation for accurate analysis of energy data. Then, Light gradient boosting machine(LightGBM) is used to construct nonlinear features, which are combined with strong correlation linear features to rank the importance of features, and accurately explore the correlation between wind power output and multidimensional factors. Finally, the multi-dimensional features are fused using a Transformer with an Encoder-Decoder architecture to generate a low-error power prediction sequence. The experimental results show that compared with multiple classical baseline models such as random forest, the mean value of Root Mean Squared Error (mRMSE) and mean value of Mean Absolute Error(mMAE) of this model are reduced by 61%, and the R² value is more than 98%, which has significant advantages in prediction performance. In addition, CEEMDAN, LightGBM and Transformer have shown good synergy in enhancing the performance of the model, which can effectively cope with the complex dynamic operating characteristics of wind power systems. The accurate prediction model can effectively stabilize the fluctuation risk of energy system caused by wind power integration, and provide reliable technical support for the grid dispatching of new power system.

Authors

Institutions

Publication Details

Journal
EAI Endorsed Transactions on Energy Web
Published
2026-11-16
DOI
https://doi.org/10.4108/ew.14761
Primary Topic
Energy Load and Power Forecasting
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A wind power prediction method based on CEEMDAN- LightGBM-Transformer

Chu Xiao-rui, Yangdi Xu, Jingsong ZHANG, Yuanyuan Yang et al.
6 links
EAI Endorsed Transactions on Energy Web
Energy Load and Power Forecasting
article

A wind power prediction method based on CEEMDAN- LightGBM-Transformer

Chu Xiao-rui, Yangdi Xu, Jingsong ZHANG, Yuanyuan Yang, Yukai Zhou
article en

Abstract

In the development of green energy, wind power, as the core clean energy, has the inherent intermittency, volatility and randomness of its output, which is very easy to cause the power imbalance of the energy system and the surge of grid accommodation pressure, seriously restricting the safe and stable operation of the energy system. This study proposes the CEEMDAN-LightGBM-Transformer model. First, Complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) decomposes the data into multiple intrinsic mode function(IMF), and uses the "drop_k" strategy to select IMF components with prominent feature information and low redundancy, effectively stripping the noise and invalid fluctuations in wind power output data, and consolidating the foundation for accurate analysis of energy data. Then, Light gradient boosting machine(LightGBM) is used to construct nonlinear features, which are combined with strong correlation linear features to rank the importance of features, and accurately explore the correlation between wind power output and multidimensional factors. Finally, the multi-dimensional features are fused using a Transformer with an Encoder-Decoder architecture to generate a low-error power prediction sequence. The experimental results show that compared with multiple classical baseline models such as random forest, the mean value of Root Mean Squared Error (mRMSE) and mean value of Mean Absolute Error(mMAE) of this model are reduced by 61%, and the R² value is more than 98%, which has significant advantages in prediction performance. In addition, CEEMDAN, LightGBM and Transformer have shown good synergy in enhancing the performance of the model, which can effectively cope with the complex dynamic operating characteristics of wind power systems. The accurate prediction model can effectively stabilize the fluctuation risk of energy system caused by wind power integration, and provide reliable technical support for the grid dispatching of new power system.

EAI Endorsed Transactions on Energy WebVol. 13
Xichang University (CN)
Affordable and clean energy, Climate action, Industry, innovation and infrastructure
Energy Load and Power Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon