Multivariate KDE Prediction of Wind and Solar Power Considering Spatial and Temporal Correlations

Traditional wind and solar power output forecasts often predict wind and photovoltaic generation separately, neglecting the potential correlations between them. In fact, wind and solar generation in the same region during the same time period exhibit characteristics of correlation. Accurately predicting wind and solar power output is a crucial foundation for improving the reliability and cost-effectiveness of high-proportion renewable energy power system dispatch. To address this issue, this paper proposes a joint prediction method for wind and solar power output based on multivariate non-parametric kernel density estimation. Using historical data, we perform bivariate kernel density fitting for wind and photovoltaic power output forecast distributions, resulting in a joint prediction distribution model for wind and solar power. Compared to traditional independent prediction methods that ignore wind-solar correlations and Copula-based joint prediction methods, the proposed approach offers higher accuracy in wind and solar power output prediction. It holds significant importance for the efficient utilization of renewable energy and reducing the instability of power systems.