To address the challenge of accurate forecasting posed by the randomness and volatility of wind power output

this paper proposes a hybrid prediction model named CEEMDAN-CPO-TCN-Transformer

based on a decomposition-optimization-deep learning integration framework. The model combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)

the crested porcupine optimizer (CPO)

temporal convolutional network (TCN)

and Transformer. First

the original power series is decomposed into multiple intrinsic mode functions (IMFs) using CEEMDAN. These IMFs are then reconstructed into high

medium

and low-frequency subsequences by integrating sample entropy (SE) and K-means clustering

effectively mitigating non-stationarity. Subsequently

the CPO algorithm is utilized to optimize the hyperparameters of TCN and Transformer

where TCN captures multi-scale local features and Transformer models global long-term dependencies. The final wind power prediction is obtained by aggregating the forecasted results of all subsequences. Experimental results demonstrate that the proposed model significantly outperforms baseline models such as TCN

Transformer

and VMD-TCN-Transformer in terms of mean absolute error (MAE)

root mean square error (RMSE) and coefficient of determination (R2)

exhibiting superior prediction accuracy and generalization capability.