A Renewable Power Scenario Generation Method Based on Wasserstein Distance for Conditional Generative Adversarial Networks

The growing share of renewable energy in generation poses major challenges to power system stability. Addressing the uncertainty and volatility inherent in renewable power is crucial for overcoming these challenges. This study introduces a novel scenario synthesis technique merging Wasserstein GANs, gradient regularization, and Conditional GANs (CGAN). Specifically, this approach involves designing the network architecture for both the generator and discriminator, utilizing Wasserstein distance augmented with a gradient penalty term as the loss function to ensure stable CGAN training. The generator models the function translating noise to realistic scenarios, contingent on projected states. We validate the proposed method using real-world renewable power data and forecast data. Results indicate that our method effectively captures the uncertainty and volatility associated with renewable power.