Controllable photovoltaic scenario generation based on conditional diffusion models

Renewable scenario generation is critical in power system planning and operation, with the increasing penetration of renewable energy resources. In this paper, we propose a deep-learning-based photovoltaic scenario generation method based on Denoising Diffusion Probabilistic Models (DDPM). The proposed method includes a forward diffusion process that gradually adds Gaussian noise to the actual data, coupled with a reverse diffusion process to generate new data given noise inputs. Compared with the popular GAN-based methods, the proposed method can generate high-fidelity, diversified scenarios without mode collapse issue. Furthermore, we extend the DDPM model to a conditional DDPM (cDDPM) by incorporating attention mechanism to capture the PV-weather dependency and achieve controllable scenario generation. Case study based on real-world PV dataset demonstrate the superiority of the proposed method.