SVM和神经网络在电能质量扰动分类应用中的对比  

Comparison Study on Classification of Power Quality Disturbances with SVM and Neural Network

作　　者：秦业[1] 袁海斌[1] 袁海文[1] 崔勇[1] 王秋生[1] 

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100191

出　　处：《南京航空航天大学学报》2011年第B07期74-78,共5页Journal of Nanjing University of Aeronautics & Astronautics

基　　金：国家自然科学基金(60974058)资助项目;航空基金(2009ZD51041)资助项目;中央高校基本科研业务费专项资金资助项目;领航创新基金计划(YWF-10-01-B25)资助项目

摘　　要：提出了一种电能质量动态扰动特征向量的提取方法,分析比较了多种分类器对电能质量动态扰动的分类能力。首先采用小波包分解算法对电能质量信号某一频段内的信息进行精细分解从而提取出特征向量,然后针对该特征向量构造了相应的BP神经网络、学习向量量化(Learning vector quantization,LVQ)神经网络、自组织特征映射(Self-organizing map,SOM)神经网络及支持向量机(Support vector machine,SVM)分类器,并模拟实际电网中的复杂扰动信号提取其特征向量集,对多种分类器的分类能力进行对比。仿真结果表明,在较复杂的电能质量扰动情况中,支持向量机分类器仍能实现对信号的精确分类,对电能质量监测具有很好的应用价值。A new method for feature vector extraction of power quality disturbance is presented, and the classification ability of multiple classifier for power quality disturbance is analyzed and compared. Firstly, the wavelet packet decomposition of a band of power quality signals is used to extract the feature vector. Then, corresponding BP neural network, learning vector quantization （LVQ） neural network, selforganizing maps （SOM） neural network, and support vector machine （SVM） classifier are established. Finally, the complex disturbance wave in power grid is simulated, their feature vectors are extracted and the classification ability of multiple classifiers is compared. Simulation results show that with the complex power quality disturbance, SVM classifier can achieve the accurate signal classification of power quality monitoring.

关 键 词：小波包 电能质量 特征向量 神经网络 支持向量机 

分 类 号：TM72[电气工程—电力系统及自动化]