基于支持向量回归机的谐波分析  

Harmonics analysis based on support vector regression machine

作　　者：郑连清[1] 何立新[1] 

机构地区：[1]重庆大学电气工程学院高电压与电工新技术教育部重点实验室,重庆400044

出　　处：《电力自动化设备》2008年第2期73-75,共3页Electric Power Automation Equipment

摘　　要：传统谐波分析方法中,窄带滤波器选频法对元件参数十分敏感,受外界环境影响较大,且精度难以保证;快速傅里叶分解法会产生频谱泄漏误差,不同频率谐波之间也有干扰;神经网络分析方法检测精度高,但其基于大样本的学习方法,受样本及训练初始值影响较大,有时会导致训练不成功。为减小外界环境影响,提高检测精度和降低噪声影响,提出基于支持向量回归机的谐波分析方法,它是基于统计学习理论,以结构风险最小化为原则的机器学习,通过引入松弛变量和损失函数提高算法泛化能力和减小误差,该算法最终转化为标准二次规划问题,有全局最优解。通过算例分析,并与傅里叶检测方法相比较,可知该算法具有稳定性好、检测精度高、对噪声不敏感等优点。In traditional harmonics analysis methods,narrow- band truer frequency selecuon method can not ensure its accuracy as it is sensitive to component parameters and environment;Fourier detection method may bring spectrum leakage and interferences among different frequency harmonics; neural network analysis method is based on large sample study,it has higher detection accuracy but depends greatly on ＇the samples and initial training value,sometimes training failure may occur. A harmonics analysis method based on SVR（Support Vector Regression machine） is presented,which is based on statistic learning with the principle of structural risk minimization. The slack variable and lose function are introduced to improve generalization capability and reduce error. It comes down to a standard quadratic programming problem and has the globally optimal solution. A Case study is performed and results show that,compared with Fourier detection method,this method has better stability and precision, insensitive to noises.

关 键 词：支持向量回归机 核函数 谐波分析 二次规划 

分 类 号：TM714[电气工程—电力系统及自动化]