基于SVMD-KPCA-BiGRU-Transformer的光伏发电功率预测研究  

Research on Photovoltaic Power Generation Prediction Based on SVMD-KPCA-BiGRU-Transformer

作　　者：韩方杰 HAN Fangjie(School of Control and Computer Engineering,North China Electric Power University,Beijing 100096,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京100096

出　　处：《东莞理工学院学报》2025年第1期57-67,134,共12页Journal of Dongguan University of Technology

摘　　要：聚焦于光伏功率预测在电网调度、安全运维及系统稳定性中的核心作用,提出SVMD-KPCA-BiGRU-Transformer预测模型。该模型首先运用连续变分态分解(SVMD)技术,将光伏功率的五大关键环境变量序列细化为多模态分量,简化数据复杂性和非平稳特性。随后,采用核主成分分析(KPCA)提取并筛选有效特征,优化数据质量,提升模型学习效能。进而,模型融合双向门控循环单元(BiGRU)与Transformer网络,前者双向学习序列数据,捕捉短期动态变化;后者借助注意力机制,强化长距离依赖捕捉能力,共同提升模型对复杂光伏功率特征的解析力。实验验证,该复合模型在预测精度上超越单一Transformer及BiGRU-Transformer模型,为光伏高效并网与电力系统智能化调度提供坚实技术基础,兼具理论深度与实用价值。This article focuses on the core role of photovoltaic power prediction in power grid scheduling,safe operation and maintenance,and system stability,and innovatively proposes the SVMD-KPCA BiGRU Transformer prediction model.The model first applies the Continuous Variational Decomposition(SVMD)technique to refine the sequence of the five key environmental variables of photovoltaic power into multimodal components,simplifying data complexity and non-stationary characteristics.Subsequently,kernel principal component analysis(KPCA)was used to extract and screen effective features,optimize data quality,and improve model learning efficiency.Furthermore,the model integrates a bidirectional gated recurrent unit(BiGRU)and a Transformer network,where the former learns sequence data bidirectionally and captures short-term dynamic changes;the latter utilizes attention mechanisms to enhance the ability to capture long-range dependencies and jointly improve the analytical power of the model for complex photovoltaic power characteristics.Experimental verification shows that the composite model outperforms single Transformer and BiGRU Transformer models in terms of prediction accuracy,providing a solid technical foundation for efficient grid connection of photovoltaics and intelligent scheduling of power systems,with both theoretical depth and practical value.

关 键 词：光伏发电功率预测 核主成分分析 连续变分模态分解 BiGRU-Transformer模型 

分 类 号：TK519[动力工程及工程热物理—热能工程]