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 can provide comprehensive information for power system scheduling and is of great significance for the operation of distribution networks. Addressing the issue of limited consideration of the dependency between photovoltaic power forecasting results and forecast errors in existing research, this paper proposes a DPV ultra-short-term power probabilistic forecasting method based on joint distribution modeling. This method effectively enhances the accuracy of probabilistic forecasting. Firstly, deterministic point forecasting of ultra-short-term power are obtained using an LSTM neural network, resulting in single-point forecasting results and forecasting errors. Then, based on these point forecasting results and errors, Copula function is used to fit the cumulative probability distribution of the point forecasting and errors, thereby constructing a joint probability distribution. Subsequently, based on the desired confidence level, upper and lower bounds of the forecasting errors corresponding to the point forecasting are obtained and added to the point forecasting, enabling the achievement of DPV ultra-short-term probabilistic forecasting. Finally, the superiority of the proposed method is validated through experiments with real-world datasets.