Scenario Generation of PV Power Based on DDPM

Along with increasing penetration of renewable energy sources, such as photovoltaics, it is important to model the uncertainty of PV power output. Scenario generation is a major method to solve the problem. This paper proposes a novel model based on data driven method for PV scenario generation based on denoising diffusion probabilistic models (DDPM). The method combines the Euclidean distance and L2 regularization loss function as a measure of the distribution difference between the generated noise and the original noise, and designs a UNet network structure suitable for photovoltaic scenario generation. In order to capture the fluctuation characteristic and achieve the efficient generation, it continuously adds noise to the forward process and constantly denoises to the reverse process. Compared with existing data driven method, such as generative adversarial networks (GAN) and variational autoencoder (VAE), the simulation results demonstrate that, the proposed model can generate PV power scenarios effectively.