Improved Conditional Generative Adversarial Network-based Approach for Extreme Scenario Generation in Renewable Energy Sources

With the gradual increase in grid size and the utilization of an increasing number of renewable energy sources, scenario data have become increasingly important for grid operation. However, the amount of scenario data available for extreme weather events is insufficient to support subsequent work. To address this problem, this paper proposes an extreme scenario generation method for renewable energy based on a conditional generative adversarial network. This method adopts the Wasserstein distance as the loss function for the discriminator and designs a network structure suitable for generating extreme scenarios for renewable energy, and it enables the generator to learn the mapping relationship distributed among scenarios under three extreme weather conditions through game training in the context of the generative adversarial network. The proposed method is tested using real photovoltaic data and validated by comparing it with the test data distribution. The results indicate that the proposed model can more accurately describe photovoltaic power generation data under extreme weather conditions.