基于TVFEMD-SE和YJQRG的短期风电功率多步概率密度预测

Aiming at the problem of short-term wind power multi-step prediction, a multi-step probability density prediction model is proposed based on the existing probability prediction methods, which is composed of time varying filter-based empirical mode decomposition (TVFEMD), sample entropy (SE), Yeo-Johnson transformation quantile regression (YJQR) and Gaussian kernel function. In this method, the original wind power is decomposed into a series of relatively stable components by using TVFEMD decomposition technique, and then the SE theory is applied to superimpose approximate components to reduce the task load. After that, a YJQR model is established for each reconstructed component to perform 4-step wind power prediction. The parameters of the model are comprehensively optimized by grid search to achieve the best prediction performance. Finally, the quantile prediction values of different components under each quantile are accumulated and used as the input variables of Gaussian kernel function to achieve multi-step probability density prediction of wind power. Taking the wind power data set of electrician mathematical contest in modeling (EMCM) in 2011 as an example, the results show that the proposed method achieves better multi-step prediction effects in terms of accuracy, uncertainty and reliability while guaranteeing noncrossing of quantiles.