Wind power prediction method based on multi-scale feature dimension reduction and multi-view iTransformer

In order to overcome the redundancy of highdimensional meteorological features in wind power prediction and the limitations of traditional single coding structure in cross-time scale feature learning, this paper proposes a wind power prediction method based on multi-scale feature dimension reduction and multi-view iTransformer. Firstly, the wavelet transform is used to decompose the data, and the low-dimensional feature space is constructed by combining the principal component analysis method to effectively eliminate the multivariate coupling noise. Secondly, the iTransformer is improved based on the multi-view strategy, and the dual-channel feature extraction mechanism is introduced to deeply mine the multi-scale features of time series data. Finally, models such as Transformer, Informer and LightTS are selected on the quarterly wind power data set for comparative experiments and ablation experiments. The experimental results show that the proposed method improves the RMSE, MAE and $\mathbf{R}^{2}$ indexes by $\mathbf{1 6. 4 3} \boldsymbol{\%}$, 25.11% and 1.09% respectively, which significantly improves the accuracy of wind power prediction and verifies the effectiveness of the proposed method in wind power prediction task.