PV-Diff: Probability Prediction of Short-term Photovoltaic Power Interval Based on Conditional Diffusion Model

In order to reduce the impact of photovoltaic power generation uncertainty on power system, a short-term photovoltaic power interval probability prediction method based on conditional diffusion model is proposed. Firstly, the network architecture of diffusion model based on Time U-Net is constructed, which can deeply learn the ability of noise analysis and conditional guidance. Secondly, the reverse diffusion process guided by conditional enhancement is constructed to make the data generated by the model have higher accuracy. Finally, on the basis of generating multiple sets of data, the probability prediction results of photovoltaic power interval are obtained by using kernel density estimation method. Experiments are carried out on the actual data set of Australian Desert Knowledge Solar Center, and the results show that the proposed method has higher probability prediction performance in sunny, cloudy and cloudy weather conditions.