Wind Power Prediction Based on a Hybrid Model of ICEEMDAN and ModernTCN-Informer
Reliable and accurate wind power forecasting serves as one of the effective measures to enhance grid peak regulation capacity while improving the safety and stability of power systems. However, wind power generation exhibits strong randomness and volatility, which pose significant challenges to achi...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11127017/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Reliable and accurate wind power forecasting serves as one of the effective measures to enhance grid peak regulation capacity while improving the safety and stability of power systems. However, wind power generation exhibits strong randomness and volatility, which pose significant challenges to achieving precise predictions. This paper proposes a hybrid forecasting model based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) combined with ModernTCN-Informer. At first, the original wind power data undergoes ICEEMDAN decomposition to obtain several relatively stable subsequences, thereby mitigating data fluctuations. This process facilitates feature extraction from the sequences and enables the construction of a highly adaptive forecasting model. Secondly, the ModernTCN model is used to sequentially extract correlations among univariate patch sequences across multiple time steps in the dataset, long-term dependencies within univariate patch sequences, and latent correlations across variables. This effectively captures potential interrelationships in wind power data from both temporal and spatial dimensions, followed by accurate and efficient predictions using the Informer model. Finally, validation is conducted on real wind farm data, and the results show that: compared with Informer, the MAE of 12-step prediction is reduced by 3.1%, and compared with LSTM, it is reduced by 29.5%; after incorporating ICEEMDAN, the MAE is further reduced by 64.1%, and the R2 reaches 0.969. The multi-step prediction accuracy is superior to that of the comparison models, verifying the effectiveness of the proposed model. |
|---|---|
| ISSN: | 2169-3536 |