Wind Power Scenario Generation Based on Temporal Conditional Generative Adversarial Network

The uncertainty of wind power output poses significant challenges to the stable operation of power systems. Artificial intelligence-based scenario generation approach offers a new way to frame uncertainty in wind power. However, generative adversarial networks (GANs) face challenges, including limited ability to effectively capture temporal information, instability during the training process, and susceptibility to mode collapse. To this end, this paper proposes a Temporal Conditional Generative Adversarial Network (TC-GAN) for renewable energy output generation. Temporal convolutional networks are embedded in both the generator and discriminator of conditional generative adversarial networks (CGAN), with the Wasserstein distance serving as the loss function for the discriminator. This design enables adversarial training to effectively capture the mapping relationships between noise and scenarios in day-ahead forecasting conditions. Furthermore, a comprehensive evaluation metric system for generated scenarios was developed, employing statistical analysis methods to assess scenario quality. Case studies demonstrate that the proposed model effectively captures temporal dependencies and enhances the accuracy of wind power scenario generation.