Redesigning the Decoder and Loss Function of Diffusion Transformer for PV Temporal Simulation

Photovoltaic (PV) temporal simulation is a core technology for PV planning analysis. This paper proposes the Weather Diffusion Transformer (Weather-DiT), which is based on a diffusion model architecture, to achieve stochastic time series simulation of direct sunlight, scattered solar radiation, and environmental temperature (the three weather factors affecting PV power generation). Using 63 years of historical weather data in Guangdong and Beijing, China, the proposed model is validated to outperform probabilistic models and Generative Adversarial Networks in terms of probability, temporal consistency, and diversity in PV stochastic simulation. Case studies verify the proposed algorithm’s accuracy in characterizing weather trends, seasonality, random fluctuations, computational time, and performance on different datasets.