500kV自耦变压器直流偏磁振动特征提取与模式识别方法研究

Vibration Feature Extraction and Pattern Recognition Method for 500kV Autotransformer in DC Bias Condition

为了对自耦变压器直流偏磁进行有效检测,以变压器振动加速度信号为观测量,对比分析了直流偏磁前后500kV自耦变压器振动加速度频谱的变化特性,提出了奇偶次谐波比、频谱复杂度以及小波包能量等原始特征参数,利用主成分分析方法对原始特征参数去相关处理,采用最小二乘支持向量机的方法对直流偏磁主特征进行了模式识别。结果表明,直流偏磁对变压器振动时频特性影响显著,采取中性点电容隔直措施后,自耦变压器仍然存在较为严重的直流偏磁问题,所提出的原始特征参数能够有效反映变压器直流偏磁状态,利用主成分分析方法显著降低了特征空间维数,采用最小二乘支持向量机的方法能够实现自耦变压器直流偏磁主特征的模式识别,通过核函数参数对调整,识别准确率可达100%。所提出的方法为变压器直流偏磁的有效检测提供了技术支持。

The objective of this paper is to detect DC bias in autotransformers with vibration acceleration tests.Vibration spectral features of a 500 kV autotransformer before and after DC bias are compared and analyzed.Oddeven rate,spectral complexity and wavelet package energy are taken as the feature parameters of DC bias.The principal component analysis(PCA)method is used for decorrelation of these original parameters.After this process,the least squaressupport vector machine(LS-SVM)method is employed for pattern recognition of the vibration features of DC bias.It is found that DC bias exerts a large influence on transformer vibration.Even if the neutral point of the autotransformer is connected with a power capacitor,DC bias still cannot be eliminated.The proposed parameters are verified effective to detect DC bias variation in transformers.With the PCA method,the dimensions of characteristic space can be decreased to a large degree.The DC bias recognition accuracy of the proposed LS-SVM method may reach 100%with proper kernel function parameters.The proposed methodology may provide technical support for DC bias detection in autotransformers.