Ultra-short-term Power Probabilistic Forecasting Method for Distributed Photovoltaic based on Joint Distribution Modeling

Accurate probabilistic forecasting of distributed photovoltaic (DPV) ultra-short-term power is of great significance for the operation and scheduling of distribution networks. To address the issue that existing research works don't effectively consider the dependency between power forecasting errors and the fluctuation characteristics, this paper proposes a probabilistic forecasting method for ultra-short-term DPV power based on joint distribution modeling. In this method, drawing inspiration from the design concepts of Large Language Models (LLM), a Transformer model is constructed to excel in deterministic point forecasting for time series under few-shot conditions, providing high-quality data support for probabilistic forecasting. Meanwhile, to achieve the finer mining of the relationship between power forecasting errors and fluctuation characteristics, the deep uncertainty extracted and measured with Autoencoder (AE) is designed to support effectively recognition of power fluctuation patterns. Finally, the joint probability distribution between point forecasting values and errors under different fluctuation patterns is fitted based on Copula function, and the upper and lower limits of the forecasting errors corresponding to the point forecasting values are obtained through inverse transformation, forming the forecasting interval. The superiority of the proposed method is validated through real datasets