摘要
针对大规模光伏并网给电力系统安全稳定运行带来的严峻挑战,考虑传统单一光伏场站功率预测的局限性,以区域性光伏集群功率为研究对象,提出一种基于BP神经网络的光伏集群功率的区间预测方法。通过互信息方法对变量进行相关性分析,提取关键解释变量作为输入变量,利用主成分分析进行数据降维,解决了光伏集群功率预测大数据处理的问题。利用神经网络在数据挖掘和非线性关系拟合方面的优越性,将神经网络和非参数概率预测相结合,量化光伏集群功率预测结果的不确定性。实验算例采用中国某地区10个光伏场站,利用未降维的原始数据与本研究所提出的数据降维方法进行对比,分别计算80%和90%预测区间,结果表明,本研究所提出的预测方法预测区间带更窄,具有更好的预测效果。利用本研究所提模型预测了某天超前72 h的80%和90%置信区间,验证了该方法的可行性和先进性。
High penetration photovoltaic power brought severe challenges to the sage and stable operation of power systems.Considering the limitations of individual photovoltaic power prediction,this paper proposed an interval prediction method of regional photovoltaic power based on BP neural network,which extracted correlation through mutual information,and applied principal component analysis(PCA)to data dimensionality reduction.Taking the advantages of the data mining and nonlinear relation fitting for BP neural network,the uncertainty of regional photovoltaic power prediction could be quantified by the combining the neural networks and nonparametric probabilistic prediction methods.Experimental example used 10 photovoltaic plants in an area in China and compared the method of no data dimension reduction of original data with the proposed method in this paper.The 80%and 90%confidence intervals of the two models were calculated,which showed that the confidence intervals of the proposed mothed were narrower.At the same time,the proposed model predicted the 80%and 90%confidence intervals with 72 hours in advance,the results verified the feasibility and advance of the proposed method.
作者
孙东磊
王艳
于一潇
韩学山
杨明
闫芳晴
SUN Donglei;WANG Yan;YU Yixiao;HAN Xueshan;YANG Ming;YAN Fangqing(Economic and Technology Research Institute,State Grid Shandong Electric Power Company,Jinan 250021,Shandong,China;School of Electrical Engineering,Shandong University,Jinan 250061,Shandong,China)
出处
《山东大学学报（工学版）》
CAS
CSCD
北大核心
2020年第5期70-76,共7页
Journal of Shandong University（Engineering Science）
关键词
光伏集群
区间预测
BP神经网络
数据降维
主成分分析
regional photovoltaic power
prediction intervals
BP neural network
dimensionality reduction
principal component analysis
