面向高压断路器故障分类的电流-振动信号类聚几何敏感特征优选方法

Optimization Method of Clustering Geometric Sensitive Features of Current Vibration Signals for Fault Classification of High Voltage Circuit Breakers

针对利用电流-振动信号联合进行高压断路器故障分类过程中原始特征集维数较高,因而造成识别率低、分类性能退化的问题,该文提出一种适用于高维小样本的类聚几何敏感特征优选方法。首先,利用广义维数谱和敏感维数定量刻画经魏格纳-威尔分布处理的振动信号时频图,结合线圈电流信号的突变信息,构建断路器完整动作过程的电-振联合多征兆域原始特征集;其次,通过定义“变异系数”细致描绘特征在样本类内和类间的波动性,由“赏函数”对类间发散性强的特征予以加权,根据特征敏感因子优选得到不同故障类型下的类聚几何最优特征集;最后,采用支持向量机和其他识别方法进行故障分类。试验结果表明采用最优特征集识别准确率明显提高,具有工程应用价值。

High voltage circuit breaker is the key safety control equipment of power system,and the loss caused by its failure is far more than its own value.The action of the circuit breaker involves the secondary electrical circuit control and the energy transmission process between mechanical components.Its complex structure and harsh operating environment are easy to cause electrical or mechanical failures.The control coil current and the vibration signal in the transmission process are effective characteristics for analyzing the abnormal operating state of the circuit breaker.In the process of fault classification of high-voltage circuit breakers using current vibration signals,the dimension of the original feature set is high,resulting in low recognition rate and degradation of classification performance.This paper proposes a clustering geometric sensitive feature optimization method suitable for high-dimensional and small samples.Firstly,the time-frequency diagram of vibration signal processed by Wigner-Ville Distribution(WVD)is quantitatively characterized by generalized dimension spectrum and sensitive dimension.The time-frequency diagram reflects the energy difference of the circuit breaker vibration signal in the form of color scale.The generalized dimension spectrum and sensitive dimension quantitatively depict the local characteristics of the timefrequency diagram,and carefully capture the time-frequency changes.Secondly,using the mutation information of the coil current signal decomposed by singular value decomposition to extend the time history of the vibration signal,the original feature set of the electrical vibration joint multi symptom domain for the complete action process of the circuit breaker is constructed.Finally,the"coefficient of variation"is defined to describe the volatility of the characteristics in the samples within and between classes,the"reward function"is used to weight the characteristics of strong divergence between classes,and the optimal collection of clustering geometry under different fault types is obtained according to the optimization of feature sensitivity factors.The corresponding relationship between the fault types and the optimal feature set is clarified.Taking ZN63-12 high-voltage circuit breaker as the research object,a fault simulation experiment platform is built.The experimental platform is used to simulate nine states of the circuit breaker,including normal,electrical fault,mechanical fault and compound fault.For each type of target state group,the feature sensitivity factor of a specific state group is calculated respectively,and the support vector machine(SVM)method is used to classify faults,and finally the optimal feature set under a specific fault type is selected.The experiment shows that:①The original feature set of electrical vibration combined multi symptom domain in the complete action process of the circuit breaker can fully reflect the electrical fault,mechanical fault and compound fault of the circuit breaker.②The optimization method of clustered geometric sensitive features is proposed to solve the problem of highdimensional small samples of fault classification,and the optimal feature set for different fault types is scientifically screened through more comprehensive"rewards and punishments".③The SVM method based on clustering geometric optimization feature set gives consideration to the accuracy of fault classification and computational performance,and has engineering application value.