基于改进支持向量机的电能质量扰动分类

Power quality disturbance classification based on improved support vector machine

对近年来电网发展和研究的热门话题之一:电能质量扰动识别分类系统进行研究。识别分类系统使用小波变换方法对扰动电压信号进行特征提取,之后收入由支持向量机建立的识别系统中。相对小波能量只能表达总分解层信号能量中各层信号能量的比例,对于电能质量扰动信号的检测不能直接使用信息熵公式。因此引入加权算子以改进相对小波能量,加权算子对电能扰动特征进行放大,实时反映电能扰动特征。针对使用支持向量机建立电能质量扰动识别系统时会由于扰动信号特征向量维度高、数据庞大等问题,提出一种基于混合核函数的LSSVM建立电能质量扰动识别系统。选取RBF核函数和Polynomial核函数分别作为局部以及全局核函数,构造混合核函数,提高系统泛化能力。使用PSO优化算法对LSSVM分类器进行分类,提高分类器的识别精度等性能。最后通过实验验证研究的电能质量扰动识别分类系统的识别性能。

The disturbance identification and classification system of power quality is studied,which is one of the hot topics of power grid development and research. The identification and classification system uses the wavelet transform method to extract the features of the disturbance voltage signal,and then transmits the features to the recognition system established by support vector machine. Since the relative wavelet energy can only express the proportion of the signal energy of each layer in the signal energy of the total decomposition layer,the detection of power quality disturbance signal can＇t directly use the information entropy formula. The weighted operator is introduced to improve the relative wavelet energy,which can amplify and reflect the power disturbance features in real time. Since the feature vector of disturbance signal has the problems of high dimension and large data when the support vector machine is used to establish the power quality disturbance identification system,a method of using LSSVM based on hybrid kernel function to establish the power quality disturbance classification system is proposed.The RBF kernel function and Polynomial kernel function are selected as the local and global kernel functions respectively to construct the hybrid kernel function. The hybrid kernel function can improve the generalization ability of the system. The PSO algorithm is used to classify the LSSVM classifiers and improve the recognition accuracy and performance of the classifiers. The recognition performance of the power quality disturbance identification and classification system was verified in experiments.