Wind Power Scenario Generation Based on Dual‐Domain Generator and Improved Spectral Bias Generative Adversarial Network

Scene generation is a core technology for quantifying wind power uncertainty. Addressing issues such as insufficient scene realism and low spectral matching accuracy in traditional generation models, this paper proposes a frequency domain generative adversarial network (FD‐GAN) method for wind power scene generation based on a dual‐domain generator and improved spectral deviation. First, a frequency domain branch is added to the original time domain main branch of the GAN. This branch performs a Fourier transform on the input signal to extract its frequency domain features. The two branches are then fused, maintaining the existing stability while incorporating frequency domain correction capabilities. Second, it couples adversarial learning with frequency‐domain loss by introducing a frequency‐domain alignment loss to the original WGAN‐GP objective function. This compares spectral differences between generated and real data, dynamically focusing on error‐prone frequency bands to enhance optimization efficiency. Experimental results demonstrate that this method provides high‐precision scenario support for quantifying wind power uncertainty, thereby providing a reliable data foundation for subsequent power system planning and dispatch decisions. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.