基于DBSCAN的电力变压器故障诊断

Fault diagnosis of power transformer based on DBSCAN

三比值法是目前电力系统中发现潜伏性故障行之有效的方法之一,但其存在编码缺码和编码边界过于绝对等不足之处,鉴于此,利用具有噪声的密度聚类(density-based spatial clustering of applications with noise, DBSCAN)算法建立了电力变压器故障诊断模型,将多种故障类型的典型油色谱数据作为模型的输入得到多种故障的典型聚类簇,并用测试数据对其进行验证。结果表明该方法能在一定程度上解决三比值法的缺码问题,诊断出一些三比值法无法诊断的故障情况,填补了三比值法的不足;同时,对于一些边界值数据,因该模型产生的聚类簇的边界可以形成任意形状,故而能很大程度上克服三比值法在编码边界误判的缺陷,为油中溶解气体分析方法提供了新思路。在实际案例中利用该模型对某变电站主变进行了分析诊断,结果与实际检测情况吻合,验证了该方法的实用性。

The three-ratio method is currently one of the effective methods to find latent faults in power system,but it has shortcomings such as codes missing and too absolute coding boundaries. In view of this, this paper uses the density-based spatial clustering of applications with noise(DBSCAN) to establish a power transformer fault diagnosis model. The typical oil chromatogram data of multiple fault types are taken as the input of the model to obtain the typical cluster of various faults,and the test data are used to verify the cluster. The results show that this method can solve the problem of missing codes in the three-ratio method to a certain extent, and diagnose some fault conditions that the three-ratio method cannot diagnose. So this method can fill the deficiencies of the three-ratio method. At the same time, for some boundary value data, because the boundary of the cluster generated by this model can be formed into any shape, it can largely overcome the defect of misjudgment of the three-ratio method in the coding boundary. It provides a new idea for the analysis method of dissolved gas in oil. In the actual case application, the model is used to analyze and diagnose the main transformer of a substation, and the results are consistent with the actual detection conditions, which verifies the practicability of the method.