摘要
提出了一种基于广义学习矢量量化的负荷特征聚类方法,既考虑了专家经验的指导,同时兼备人工神经网络强大的非线性处理能力。归纳提取的负荷、气象综合指数更能反映日负荷的基本特征,同时减少了网络输入层的维数。实际预测结果表明,综合了广义学习矢量量化和支持向量机的混合短期负荷预测方法无论是在聚类准确度方面,还是在预测准确度方面,与单纯的支持向量机算法和自组织特征映射与支持向量机的混合算法相比,均具有明显的优势。
A load characteristic clustering method based on generalized learning vector quantity （GLVQ） is proposed. This method not only considers the guidance of expert experiences but also possesses powerful nonlinear processing ability of artificial neural network （ANN）. The induced load and comprehensive meteorological index can reflect the basic characteristics of daily load better; meanwhile, the dimensions of network input layer can be reduced. Practical short-term load forecasting results by the proposed method show that compared with pure support vector machine （SVM） algorithm and the hybrid algorithm of self-organized feature mapping （SOFM） and SVM, whether in clustering accuracy or in forecasting accuracy, the proposed hybrid short-term load forecasting method based on GLVQ and SVM possesses evident superiority.
出处
《电网技术》
EI
CSCD
北大核心
2008年第13期62-68,共7页
Power System Technology
关键词
短期负荷预测
负荷特征聚类
广义学习矢量量化
(GLVQ)
支持向量机(SVM)
气象因素
short-term load forecasting
load characteristic clustering
generalized learning vector quantity （GLVQ）
support vector machine （SVM）
meteorological factor
