Controllable renewable energy scenario generation based on pattern-guided diffusion models

The frequent occurrence of extreme weather events attributed to global climate change presents challenges to the supply-demand balance in power systems with high renewable energy integration. Modeling renewable energy scenarios can effectively guide power system operation and planning. However, obtaining diverse extreme renewable energy scenarios is challenging due to the limited availability of renewable power dataset resources. To tackle this issue, the pattern-guided diffusion model (PGDM) is proposed for controllable renewable energy scenario generation. Initially, we define the scenario pattern features associated with wind and solar power generation. A contrastive pre-training model is used to learn representations of renewable energy scenarios, aiding the downstream model in understanding scenario pattern features. Subsequently, a perceptual variational autoencoder is used to map the high-dimensional scenario into a low-dimensional latent space, thus reducing the computational burden. This is combined with a conditional latent diffusion model to achieve controllable scenario generation. The renewable power dataset from Belgian transmission operator Elia was used for the case study. The proposed PGDM demonstrated lower errors in controllable scenario generation and exhibited excellent generalization performance in low probability (few-shot) and novel (zero-shot) pattern scenario generation.