A Deterministic and Probabilistic Prediction Method for Short- Term Photovoltaic Power Considering Spatial Correlation

Abstract

With the application of distributed photovoltaics in recent years, spatial correlation is necessary to improve forecasting accuracy. Therefore, a deterministic and probabilistic forecasting method of short-term photovoltaic power considering spatial correlation is proposed in this paper. This method firstly analyzes the spatial correlation of multiple photovoltaic sequences of similar power stations and selects the reference photovoltaic sequence with a strong correlation. Then, an approach is proposed for selecting similarity days based on grey correlation degree theory. The selected similarity days are used as a training model, and Extreme Gradient Boosting (XGBoost) is utilized to construct a prediction model to obtain single-value forecasting results. Finally, by obtaining the corresponding error probability density function and error distribution interval, and adding up to the single-value forecasting result, the photovoltaic power probability prediction can be realized. The effectiveness of the proposed method is verified by the simulation.