An Interpretable Photovoltaic Scenario Generation Method Based on RMT-Enhanced WGAN Considering Physical Constraints

Accurate generation of renewable energy scenarios is critical for both planning and scheduling in power systems. Conventional generative models for renewable power scenario generation often suffer from poor interpretability and inadequate adherence to practical physical constraints. To address these limitations, an interpretable photovoltaic (PV) scenario generation method based on random matrix theory (RMT) enhanced Wasserstein Generative Adversarial Network (WGAN) is proposed. In the proposed framework, the WGAN generator is enhanced with a dynamic gate function to ensure that the generated scenarios comply with practical physical constraints of PV outputs. Furthermore, the WGAN discriminator incorporates the ring law from random matrix theory, thereby enhancing both the interpretability of the generative model and the quality of the generated scenarios. Case studies are performed to verify the applicability and interpretability of the proposed scenario generation framework, with RMT-WGAN achieving a lower MAE of 0.012 and a higher SSIM of 0.951.