基于NMF-SVM的光伏系统发电功率短期预测模型  

Short-Term Photovoltaic Generation Forecasting System Based on NMF and SVM

作　　者：吴江[1] 卫志农[1] 李慧杰[2] 李晓露[2] Kwok W Cheung 孙永辉[1] 孙国强[1] 

机构地区：[1]河海大学可再生能源发电技术教育部工程研究中心,南京210098 [2]阿尔斯通电网技术中心有限公司,上海201114 [3]ALSTOM Grid Inc.

出　　处：《华东电力》2014年第2期330-336,共7页East China Electric Power

基　　金：国家自然科学基金项目(51277052;51107032;61104045)~~

摘　　要：根据光伏发电系统的历史发电数据和气象数据,考虑天气类型、日照强度和大气温度及风速等因素,提出一种基于非负矩阵分解(nonnegative matrix factorization,NMF)和支持向量机(support vector machine,SVM)的光伏系统发电功率短期预测模型。基于差异性和相关性原理,同时考虑相似日选择算法,通过NMF算法对由相似日组成的输入样本进行分解,得到非负的低维映射矩阵,将其作为支持向量机的输入,预测光伏系统的发电功率。该模型在消除冗余信息、减少变量维数的同时,保留了原始问题的实际意义。实例表明,该方法降维效果明显,预测精度得到显著的提高。With regard to the historical data about power generation and weather condition, as well as the influencing factors, such as weather types, sunshine intensity, temperature, wind speed, etc. , a new short-term forecasting mod- el for power output of a PV power system is proposed based on nonnegative matrix factorization （NMF） and support vector machine （SVM）. On the basis of the relevance and difference principle and the similar day selection algo- rithm, a method is proposed to select similar clays for PV array output power. The input data is decomposed by using the NMF algorithm, then the derived nonnegative mapping matrix with lower dimension is taken as the input of SVM for PV output forecasting. This model possesses some good properties such as eliminating redundant data, reducing variable dimension, etc. , and thus it could keep the practical significance of the original problem. Finally, simula- tion results are provided to show that the dimension of the input variables can be effectively reduced, and the accuracy could also be greatly improved.

关 键 词：光伏系统 非负矩阵分解 支持向量机 气象因素 相似日选择算法 发电功率预测 

分 类 号：TM615[电气工程—电力系统及自动化]