Dual-path frequency Mamba-Transformer model for wind power forecasting

Accurate wind power forecasting is fundamental to the secure and stable operation of renewable energy-dominated power systems. Current wind power forecasting faces two major challenges: 1) accurately characterizing the rapid fluctuation processes of wind power series, and 2) effectively handling the correlations among variables including wind speed, pressure, temperature, and humidity. To address the limitations of existing wind power prediction methods in extracting the high-frequency features and global information, This study proposes a dual-path frequency Mamba-Transformer (DPFMformer) model with maximal information coefficient-weighted error correction (MEC). First, The DPFMformer model decomposes wind power series through multi-scale Fourier transform, processing real components via attention mechanisms and Mamba layer, and imaginary components through multilayer perceptron (MLP) layers to utilize frequency domain information. Second, a frequency kernel loss function is proposed to enhance the DPFMformer's ability to learn high-frequency features. Finally, a MEC method is proposed to utilize the correlations among meteorological variables including wind speed, pressure, temperature, and humidity for improved forecast accuracy. The effectiveness of our approach is demonstrated through experiments on datasets from two wind farms across four seasons. The results show that the proposed model achieves significant improvements over traditional methods, with an average root mean square error reduction of 61.3 %.