关键词: 风电功率预测, CEEMDAN, 样本熵, K-means聚类, 时间卷积网络, 双向门控循环单元

Abstract: Improving the accuracy of wind power forecasting is critical to ensuring the safe and stable operation of power grids. However, due to the high randomness and volatility of wind energy, traditional forecasting methods often struggle with limited feature extraction and modeling capabilities. To address these challenges, this study proposes a novel short-term wind power forecasting model, termed CEEMDAN-DBO-VMD-TCN-BiGRU, which integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Dung Beetle Optimization (DBO), Variational Mode Decomposition (VMD), Temporal Convolutional Network (TCN), and Bidirectional Gated Recurrent Unit (BiGRU). First, CEEMDAN is employed to decompose the original wind power series into intrinsic mode functions (IMFs), capturing key temporal features. The IMFs are then classified into high-, medium-, and low-frequency components using sample entropy and K-means clustering. High-frequency components are further decomposed via DBO-optimized VMD to enhance feature representation and reduce computational complexity. All components are normalized and fed into the TCN-BiGRU hybrid network for prediction. The final forecast is obtained by aggregating the individual predicted components and applying inverse normalization. Experimental results demonstrate that the proposed model achieves a root mean square error (RMSE) of 6.47 MW, a mean absolute error of 4.30 MW, a mean absolute percentage error of 6.28%, and a coefficient of determination of 98.62%, outperforming benchmark models and confirming its effectiveness, robustness, and practical applicability.

Key words: wind power forecast, CEEMDAN, sample entropy, K-means, TCN, BiGRU