Abstract: While transient power quality disturbances are made classifications, single feature can not be fully representative of transient disturbance, it may cause errors. In order to overcome the weaknesses, this paper extracts scheme of power quality multi-classification based on combined features and binary tree support vector machine. Using wavelet packet transform to decompose the disturbance signal is extracted under specific band signal energy, and the feature information is extracted by modulus matrix of perturbation signal using S transform. Then the multi-band signal energy and the corresponding S conversion feature information grouped together to get a combination of features. Finally, the binary tree support vector machine classifier with clustering ideas is trained and tested. The simulation result shows that the proposed method can effectively classify disturbance signals with higher recognition accuracy and faster recognition speed, the average recognition rate increases more than 6%, total test time shortened 0.06 s, the training time is reduced to 1.8 s.

Key words: wavelet package transform, S transform, binary tree, support vector machine, power quality

摘要： 为了克服单一特征不能完全表征各种暂态扰动信号特征的不足，提出了一种基于组合特征和二叉树结构支持向量机相结合的电能质量多分类方案。利用小波包变换对扰动信号进行分解，提取特定频带下信号的能量，利用S变换获得扰动信号的模矩阵，从中提取出特征信息，然后将多频带信号的能量和对应的S变换特征信息组合得到组合特征。对依据聚类思想设计出的二叉树结构支持向量机分类器进行了训练和测试。仿真结果表明，该方法具有较好的准确性和识别速度，能够有效识别常见扰动信号，平均识别率提高了6%以上，测试总用时缩短0.06秒，训练时间减小1.8秒。

关键词: 小波包变换, S变换, 二叉树, 支持向量机, 电能质量