Ultra-Short-Term Distributed PV Power Forecasting Based on Correlation Sparse Modeling and Uncertainty Pattern Forecasting

Abstract

Accurate ultra-short-term power interval forecasting for distributed photovoltaic (DPV) systems is crucial for the intra-day operation of distribution networks. However, existing studies still face two major limitations. First, most spatiotemporal forecasting methods are developed for a limited number of DPV sites. When extended to large-scale DPV fleets, the rapidly increasing model complexity may lead to excessive computational burden and a higher risk of overfitting. Second, existing interval forecasting approaches generally overlook the heterogeneous uncertainty characteristics under different forecasting scenarios, limiting their ability to accurately characterize scenario-dependent forecast error distributions. To address the issues above, this paper proposes an ultra-short-term DPV power interval forecasting method based on correlation sparse modeling and uncertainty pattern forecasting. First, a correlation sparse point forecasting model is developed to identify representative DPV sites and exploit their informative inter-site dependencies, thereby enabling efficient regional power forecasting while controlling model complexity. Subsequently, uncertainty characteristics that directly reflect variations in forecast error distributions are extracted to identify distinct uncertainty patterns. For each pattern, a dedicated joint probability distribution model is constructed to achieve refined characterization of scenario-dependent forecasting uncertainty. Finally, the predicted uncertainty pattern is combined with the point forecasting result to select the corresponding probabilistic model and generate the power prediction interval. The effectiveness of the proposed method is validated using a real-world DPV dataset from China.