Wind Power Prediction of BiTCN-BiGRU-KAN Model Based on Attention Mechanism

With the advancement of renewable energy, wind power generation is increasingly becoming a vital factor in ensuring the stability of power systems. However, as an inherently fluctuating energy source, the large-scale integration of wind power into the grid can potentially amplify stability risks. Hence, accurate short-term forecasting of wind power output is essential to enhancing the reliability and precise control of grid operations. To increase the precision of wind power prediction, this paper proposes a model based on the integration of Bidirectional Temporal Convolutional Networks (BiTCN), Bidirectional Gated Recurrent Units (BiGRU), and Kolmogorov-Arnold Networks (KAN). The model initiates by employing the BiTCN to extract global spatial features, capturing spatial correlations and temporal dependencies in the data. Subsequently, the BiGRU network is utilized to extract sequential features, identifying long-term dependencies and local patterns. The introduction of attention mechanisms facilitates the fusion of spatial and sequential features, enhancing the model's performance and generalization capabilities. Lastly, the KAN model generates the final power forecast by acting as a completely connected layer. Based on experimental data, the suggested model outperforms other single and combined models in terms of precision, achieving a Mean Absolute Error (MAE) of 5.2853, a Root Mean Squared Error (RMSE) of 11.7850, and an R-Squared (R2) value of 0.9460. The model's improved ability to anticipate wind power in the short term with accuracy is validated by these indicators.