特征提取与缩减在电力变压器异常状态检测的应用

Feature extraction and reduction applied to anomalous state detection of power transformer

现有的三相干式变压器振动法状态检测,是一种利用从高频振动数据中提取特征,从而实现检测的方法。为了更准确地挖掘高频振动信号中的有效特征,提高诊断准确性,针对高频振动信号,将连续信号划分为独立的周期片段,从每一个周期中提取信号的时域、频域和时频域特征;利用变压器振动原理分析,结合相关性分析给出特征选取策略;将筛选后的特征作为故障的特征值对振动进行实时的异常检测。结果表明:该方法可有效提取变压器振动信号特征,提高诊断准确率。

The existing vibration method for three-phase dry-type transformer state detection uses feature extraction from high-frequency vibration data to achieve detection.In order to mine the effective features of high-frequency vibration signals more accurately and improve the accuracy of diagnosis,for high-frequency vibration signals,the continuous signals are divided into independent periodic segments,and the time-domain,frequency-domain and time-frequency-domain features of the signals are extracted from each cycle;Then,based on the analysis of transformer vibration principle and the correlation analysis,the feature selection strategy is given;Finally,the filtered features are used as fault eigenvalues to detect vibration anomalies in real-time.The results show that this method can effectively extract the characteristics of transformer vibration signals and improve diagnostic accuracy.