西北工程技术学报2025，Vol.24Issue(2)：118-129,12.
基于两次多元分解和筛选的风电功率预测方法
Wind Power Prediction Method Based on Two-Stage Multivariate Decomposition and Feature
摘要
Abstract
To address the challenges of data noise and the insufficient utilization of multivariate coupling relationships in wind power prediction,a hybrid prediction model is proposed based on two-stage multivariate decomposition and feature selection.The proposed method first employs Multivariate Variational Mode Decomposition(MVMD)to decompose multivariate time series data into multi-scale components,with subsequent reconstruction of each variable's sub-modes into high-,medium-,and low-frequency components using sample entropy and hierarchical clustering.A secondary MVMD decomposition is then implemented on the high-frequency components for refined feature extraction.Innovatively,a two-stage feature screening mechanism based on the Granger causality test is introduced to effectively eliminate non-stationary sequences and components that lack causal relationships.Compared to the wind power prediction results from Long Short-Term Memory(LSTM)and Convolutional Neural Networks(CNN),the proposed method can reduce the Mean Absolute Error(MAE)by 89.8%and the Root Mean Square Error(RMSE)by 90.6%,significantly outperforming univariate decomposition methods.Meanwhile,the two-stage multivariate decomposition and screening approach can significantly improve the model's predictive accuracy compared to methods that do not utilize decomposition and those that use single-stage decomposition.The Diebold-Mariano test confirms that the proposed model outperforms all benchmark models at a 5%significance level,with the improvement rate stabilizing above 27%.This study proves that the hybrid model can significantly enhance data quality,thereby improving the results of wind power prediction and providing a new solution for high-precision wind power forecasting.关键词
风电功率预测/深度学习/数据预处理/多元变分模态分解/数据筛选Key words
wind power prediction/deep learning/data preprocessing/multivariational mode decomposition/data selec-tion分类
信息技术与安全科学引用本文复制引用
李昊,胡春生,巩豪委..基于两次多元分解和筛选的风电功率预测方法[J].西北工程技术学报,2025,24(2):118-129,12.基金项目
宁夏重点研发计划项目(2024BEE02003) （2024BEE02003）
