基于核主成分分析和极限学习机的短期电力负荷预测  

Short-term power load forecasting based on kernel principal component analysis and extreme learning machine

作　　者：董浩 李明星 张淑清[3] 韩立强[3] 李军锋[3] 宿新爽 

机构地区：[1]河北省自动化研究所 [2]石家庄铁道大学 [3]燕山大学电气工程学院

出　　处：《电子测量与仪器学报》2018年第1期188-193,共6页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(61077071);河北省自然科学基金(F2015203413,F2016203496);河北省高等学校科技研究重点项目(ZD2014100)资助

摘　　要：电力负荷预测的影响因素很多,需要综合考虑多个指标。各种指标间的关系通常是非线性的,采用线性主成分分析(PCA)往往会出现各主成分的贡献率太过分散,找不到具有全面综合能力的成分的情况。核主成分分析(KPCA)作为非线性主成分评价模型,通过核技巧,规避了非线性主成分分析(NLPCA)中非线性变换的未确知性,获得的主成分的贡献率比较集中,得到的评价结果更符合客观事实。采用KPCA来改进极限学习机(ELM)神经网络的输入量,兼顾了各个指标间非线性关系,以保留大部分原始信息为前提,有效的降低了输入维数,以极限学习机为预测模型,对实际电网中的负荷数据进行预测分析,结果表明,KPCA-ELM方法有效地提高了预测精度。There are many factors that affect the load forecasting,and it is necessary to consider various indicators synthetically,but the relationship between the actual indicators are usually non-linear,while the principal component analysis( PCA) method has no obvious effects in finding a comprehensive index,and it usually results in the contribution of the principal components to be too scattered. The paper presents a nonlinear evaluation model: kernel principal component analysis( KPCA),which could solve the uncertainty of the nonlinear transformation cleverly in the process of nonlinear principal component analysis( NLPCA),and the contribution rate of the principal component is relatively concentrated,while the evaluation result obtained is in line with the objective reality. Then the kernel principal component analysis( KPCA) is used to improve the input of extreme learning machine( ELM) neural network. It reduces the dimension of forecasting indexes greatly and retains enough original information under the nonlinear relationship between the indexes.Short-term load forecasting model was established through Extreme Learning Machine( ELM),and the forecasting and analyzing for the actual power load system was conducted,and the comparative experiment results show that the method of KPCA-ELM can improve the prediction’s accuracy effectively.

关 键 词：短期负荷预测 多变量天气因子 核主成分分析 极限学习机 

分 类 号：TM715[电气工程—电力系统及自动化]