A hybrid short-term wind power point-interval prediction model based on combination of improved preprocessing methods and entropy weighted GRU quantile regression network

Accurate wind power prediction is significant to the stability of power system. Existing deterministic prediction methods unable to describe the uncertainty of wind power while both the point and probabilistic models have difficulty in tracking the abrupt changes in wind power accurately. To settle these problems, a point-interval prediction model based on combination of improved preprocessing methods and entropy weighted GRU quantile regression network (QR-EGRU) is proposed. Firstly, an improved wavelet threshold denoising (IWTD) is applied to reduce noise interference. An optimized variational mode decomposition (OVMD) based on sparrow search algorithm (SSA) is proposed to decompose the series into subsequences. Secondly, two update gate matrices based on information entropy (IE) are introduced to replace the traditional update gate matrix of the GRU to construct the EGRU. Point prediction results are obtained by using the EGRU model. Furthermore, the QR algorithm with nonlinear loss function is derived to realize the interval prediction of the EGRU. Finally, the proposed model is validated on real wind power data from the Kaggle competition. Experimental results demonstrate that the proposed model performs well in both point and interval prediction. It can track the mutation series more precisely than other models and improve the prediction accuracy.