A novel wind speed prediction method based on fractal wavelet decomposition explainable gated recurrent unit

Wind power is characterized by prominent economic benefits and broad application prospects, but the variability of wind speed can influence the efficiency and reliability of wind power generation. To enhance the accuracy of wind speed prediction, a novel wind speed prediction method based on fractal wavelet decomposition explainable gated recurrent unit neural network is proposed. Fractal wavelet transform is studied to decompose wind series into more stable subseries. The analysis on the fractal characteristics of wind series ensures the satisfactory decomposition effect of fractal wavelet transform. Then every subseries is modeled and predicted by the gated recurrent unit model, the quantile loss function is incorporated in the training process to enhance the prediction performance of gated recurrent unit model. The predicted subseries is reconstructed to obtain final predicted results for original wind series. Moreover, shapley additive explanations method is introduced to identify the marginal contribution of each subseries, thereby addressing the explainability of the proposed model. The performance of the proposed model is evaluated by analyzing four groups of wind datasets from the real wind farm. The proposed model outperforms three hybrid models in terms of prediction accuracy, demonstrating a remarkable capability of capturing the variation pattern of wind speed series.