摘要
针对电力物资需求预测中存在的模态混叠、参数敏感及非平稳序列分解精度不足等问题,提出一种基于改进变分模态分解(VMD)算法的多阶段融合预测框架.通过构建“参数优化VMD-EMD/LSTM-GWO”协同架构,将灰狼优化算法GWO嵌入VMD模态分解过程,实现正则化参数ε的自适应调整,有效解决了传统VMD在非平稳信号处理中的模态混叠等问题.实验表明,改进后的VMD算法使预测决定系数R^(2)从0.85提升至0.95,均方误差(MSE)降低50.0%,预测误差率较原始算法减少4.48%.
Aiming at the problems of modal aliasing,parameter sensitivity,and insufficient decomposition accuracy of non-stationary sequences in power material demand prediction,this paper proposes a multi-stage fusion prediction framework based on the improved variational mode decomposition(VMD)algorithm.By constructing a“parameter-optimized VMD-EMD/LSTM-GWO”collaborative architecture,the grey wolf optimization(GWO)algorithm is innovatively embedded into the VMD modal decomposition process,achieving adaptive adjustment of the regularization parameter ε and effectively solving the modal aliasing problem of traditional VMD in non-stationary signal processing.Experiments show that the improved VMD algorithm increases the prediction coefficient of determination R^(2) from 0.85 to 0.95,reduces the mean square error(MSE)by 50%,and decreases the prediction error rate by 4.48%compared with the original algorithm.
作者
李海龙
王家平
吕强
杨贺强
杨志
LI Hailong;WANG Jiaping;LV Qiang;YANG Heqiang;YANG Zhi(Zhuhai Shenneng Hongwan Electric Power Co.,Ltd.,Zhuhai 519000,China)
出处
《云南师范大学学报(自然科学版)》
2025年第1期56-61,共6页
Journal of Yunnan Normal University:Natural Sciences Edition
基金
国家自然科学基金资助项目(51776073).
关键词
电力物资需求预测
VMD
长短期记忆网络
灰狼优化算法
多模态特征融合
Power material demand prediction
VMD
Long short-term memory network
Grey wolf optimization algorithm
Multi-modal feature fusion
作者简介
李海龙(1989-),男,甘肃天水人,工程师,主要从事电厂技术管理方面研究;通信作者:王家平.E-mail:wangjiaping@sec.com.cn.
