基于决策树和支持向量机的电能质量扰动识别  

Power Quality Disturbance Identification Using Decision Tree and Support Vector Machine

作　　者：陈华丰[1] 张葛祥[1] 

机构地区：[1]西南交通大学电气工程学院,四川省成都市610031

出　　处：《电网技术》2013年第5期1272-1278,共7页Power System Technology

基　　金：国家自然科学基金项目(61170016);教育部新世纪优秀人才支持计划项目(NCET-11-0715);配套项目(SWJTU12CX008);中央高校基本科研业务费专项资金项目(SWJTU11ZT07)~~

摘　　要：提出一种新型电能质量扰动识别方法,该方法采用快速傅里叶变换(fast Fourier transform,FFT)结合动态测度法提取3种特征以及S变换提取4种特征;采用决策树和支持向量机(support vector machine,SVM)设计组合分类器。针对FFT频谱中谐波频率明显的扰动类型,采用极值点包络的动态测度法提取频谱中的主要频率点特征,结合S变换提取的特征首先将扰动类型进行初步归类,然后采用S变换的2个特征就能进行后续分类;决策树分类过程中采用SVM来区分电压暂降和中断,克服了特征阈值随信噪比(signal-to-noise ratio,SNR)变化难以确定的问题。仿真实验表明,该方法能够准确识别包含2种复合扰动在内的11种电能质量扰动信号,SNR低至20 dB时准确率仍达到96.50%;且与已有文献的分类结果对比表明,该方法准确率高,稳定性强,在低SNR条件下分类结果优势明显。A new approach to recognize power quality disturbances is proposed. Based on fast Fourier transform （FFT） combined with dynamic measure method three kinds of features in power quality disturbance signals are extracted and using S-transform four features in power quality disturbance signals are extracted, and by use of decision tree and support vector machine （SVM） a combination classifier is designed. Firstly, for disturbance types with evident harmonic frequencies in FFT spectrum the features of main frequency points in FFT spectrum are extracted by the extreme point-enveloped dynamic measure method, and combining with the feataxres extracted by S-transform, the disturbance types are preliminarily classified into several categories, and then by use of the two features extracted by S-transform the follow-up classification can be implemented. During the classification of decision tree the SVM is used to distinguish voltage sag from voltage interruption, thus the trouble that the feature thresholds, which vary with signal-to- noise ratio （SNR）, are hard to be determined can be overcome. Simulation experiments show that using the proposed method eleven power quality disturbance signals, including two kinds of compound disturbances, can be accurately recognized, and when SNR is lowered to 20 dB the recognition accuracy can still reach to 96.50%. Comparison of the obtained results with reported classification results shows that the proposed method is accurate, stable and can be utilized in environment of low SNR.

关 键 词：电能质量 扰动识别 S变换 动态测度法 支持向量机 决策树 

分 类 号：TM76[电气工程—电力系统及自动化]