电压暂降分类特征可比性、相关性及马氏距离分类法

Comparability and Relevance of Voltage Sag Classifying Features and Classification Methods Based on Mahalanobis Distance

分类特征选取和分类方法是电压暂降分类的基础。从分类学原理出发,针对不同分类目标的电压暂降分类,引入电压暂降分类特征概念,研究同类和异类暂降分类特征的可比性,提出基于费希尔判别准则的分类特征线性判别和分类特征选取方法;考虑同类和异类暂降间分类特征相关程度的不同,提出基于马氏距离的电压暂降分类方法。以短路故障、变压器空载激磁和大型感应电动机起动引起的电压暂降为分类目标,对理论样本和不同情况下实测数据的分析证明该方法正确性、可行性和良好的噪声鲁棒性,并证明,在分类过程中考虑特征属性可有效提升其分类性能。

Feature selection and classification methods are the basis of classification of voltage sag. Starting from the classification principle, according to the different classification target of voltage sag, the concept of the voltage sag classification feature was introduced. The comparability of sag of the same classes and the different classes was researched, by proposing a linear discriminant of classification features based on Fisher principle and feature selection method. Considering the different relevance of feature sets of sag between different classes, the classification method based on Mahalanobis distance was proposed. Targeting the classification of short-circuit faults, transformer no-load excitation and startup of large induction motors, analysis of simulation samples and actual measured data on different conditions verify the correctness, effectiveness and noise~ robustness of the method, and it is proved that the consideration of the attribute of features during the classification process will promote the classification ability of features effectively.