基于局部离群因子算法的变压器异常检测

Transformer Anomaly Detection Based on Local Outlier Factor Algorithm

针对传统的变压器异常检测方法存在实时性差和效率低的问题,应用主成分分析法和局部离群因子算法(Local Outlier Factor,LOF)相结合的方法设计了变压器异常检测模型。首先,利用主成分分析法对变压器电气参量数据集进行特征降维,减少特征的冗余度;然后,通过局部离群因子算法计算所有样本点的离群因子,并将离群因子与截断阈值进行比较,筛选出变压器电气参量异常的样本点;最后,采用混淆矩阵对该方法做检测性能评估。利用局部离群因子算法对变压器状态异常进行检测,其灵敏度为81.8%,特异度为87.7%,几何均值为84.7%。局部离群因子算法的异常检测效果良好,可以辅助工程人员对变压器运行状态进行实时监测。

Aiming at the problems of poor real-time performance and low efficiency in traditional transformer anomaly detection methods,a combination of principal component analysis and local outlier factor(LOF)algorithm was used to design a transformer anomaly detection model.First,the principal component analysis method was used to reduce the feature dimension of the transformer electrical parameter data set to reduce the redundancy of the feature;then,the outlier factor of all sample points was calculated by the local outlier factor algorithm,and the outlier factor was combined with the cutoff threshold.The comparison was performed to screen out the sample points with abnormal electrical parameters of the transformer;finally,the confusion matrix was used to evaluate the detection performance of the method.The local outlier factor algorithm was used to detect the abnormal state of the transformer.The sensitivity was 81.8%,the specificity was 87.7%,and the geometric mean was 84.7%.The local outlier factor algorithm has a good anomaly detection effect and can assist engineers in real-time monitoring of the transformer′s operating status.