改进模糊聚类与主成分分析下的变压器故障识别

Transformer Fault Identification Based on Improved Fuzzy Clustering and Principal Component Analysis

针对经典模糊C均值聚类(FCM)对数据进行等权划分而造成聚类结果不理想的情况,首先,采用点密度加权方式,对变压器油中溶解气体分析(DGA)数据进行处理,提高样本可分性,削弱聚类时出现的等趋势划分对聚类中心以及分类结果造成的影响。然后,以DGA故障数据聚类中心作为变压器标准故障谱。最后,利用施加惯性系数的主成分分析方法对待测样本进行故障识别。研究结果表明:通过点密度加权的FCM对DGA数据进行故障类型分类时,平均准确率比传统FCM算法提升了9.6%。利用上述方法对多组油浸式变压器进行识别,识别结果与实测信息均一致。

Aiming at the case that the classic fuzzy C-means(FCM)clustering divided the data into equal weights and caused the clustering result to be unsatisfactory,the point density weighting was used to process the dissolved gas analysis(DGA)data in transformer oil to improve sample separability and weaken clustering.The effects of isochronous trending on clustering centers and classification results were reduced during class classification.Then the DGA fault data clustering center was used as the standard fault spectrum of the transformer.Finally,the principal component analysis method with inertia coefficient was used to identify the faults to be tested.The experimental analysis shows that the average accuracy rate of classifying DGA data by point density weighted FCM is 9.6%higher than that of traditional FCM algorithm.Using the above method for multiple groups of oil-immersed transformers,the recognition results are consistent with the measured information.