Regional Wind Power Probabilistic Forecasting Based on an Improved Kernel Density Estimation, Regular Vine Copulas, and Ensemble Learning

Abstract

Reliable wind energy forecasting is crucial for the stable operation of power grids. This paper proposes a regional wind power probabilistic forecasting model comprising an improved kernel density estimation (IKDE), regular vine copulas, and ensemble learning. The IKDE is firstly used to generate the margin probability density function (PDF) of each wind farm and the KDE bandwidth is optimized via the golden-section search algorithm to obtain the best possible prediction. Then, several dependence structures are formulated by building different regular vine copulas based on multiple criteria, and all the dependence structures work together with marginal PDF to generate respective joint distribution functions. Finally, ensemble learning is applied to combine all the joint distribution functions and establish an ultimate distribution function. Furthermore, a novel multi-distribution mega-trend-diffusion (MD-MTD) with parametric optimization is proposed to improve the prediction when the data are insufficient. The results of comparative evaluations conducted on datasets from eight wind farms indicate that the proposed model outperforms existing models in wind power generation prediction. Specifically, the proposed model can reliably forecast power generation for an entire region for the next 24 h with only three months of historical data. In contrast, most benchmark models require a year of data.