PCA_RBF网络在电力负荷预测中的应用研究  

Research about Power Load Forecasting Based on PCA and RBF Neural Network

作　　者：吴建龙[1] 郭滨钊[2] 

机构地区：[1]首钢工学院机电工程系,北京100144 [2]河南省夏邑县供电局,河南夏邑476400

出　　处：《计算机仿真》2010年第11期270-273,共4页Computer Simulation

摘　　要：研究电力负荷预测问题,由于电力负荷因子间存在非线性和高度冗余,传统方法无法消除数据之间冗余和捕捉非线性特征,导致预测精度较低。为了提高电力负荷预测精度,提出一种将主成份分析(PCA)和RBF神经网络相结合的电力负荷预测方法(PCA-RBF)。首先对电力负荷高维变量数据矩阵进行标准化处理,然后利用主成分分析建立相关矩阵,计算特征值和特征向量,通过求取累计方差贡献率,对主成分作为RBF神经网络的输入进行训练预测,主成分以较少的维数包含了原高维变量所携带的大部分信息,避免过多的输入导致的精度低和训练慢的不足。采用PCA_RBF模型对某省1992-2002的电力负荷数据进行验证性测试和分析。实验结果表明,改进的PCA_RBF模型可有效降提高负荷预测精度。The paper deals with the researching on power load forecasting problems.Because there exist highly redundant features in affecting factors of power load,traditional method cannot eliminate data redundancy and prediction accuracy is very low.In order to improve the accuracy of power load forecasting,a new method is proposed based on principal component analysis（PCA） and radial basic function neural networ（RBF）.The original high dimensional data matrix is normalized to establish the correlation rnatrix for the calculation of eigenvalues,eigenvectors and accumulated contribution of variance,according to which the principal components are determined.The low dimensional dimensional principla components with most information included in the original high dimensional data set are used as the inpus of RBF for training and prediction.The forecastiong model considers all influencing factors and avoids the low precision and slow training induced by over input.Power load is tested and verified by PCA_RBF,the results show that PCA_RBF has improved the prediction accuracy significantly,compared with the reference models.

关 键 词：主成分分析 电力负荷 神经网络 预测 

分 类 号：TM743[电气工程—电力系统及自动化]