基于级联阈值决策的变压器故障模糊诊断法

Weighted Fuzzy Identification Based on Cascaded Thresholds Applied for Diagnosing Transformers Faults

为了解决最大隶属度原则不宜诊断多故障、阈值原则漏诊率和误诊率较高的问题,在提出级联阈值判断原则后构成了级联阈值决策的加权模糊识别法。该法研究各特征气体组分比值在变压器故障诊断中的不同地位,确定加权系数矩阵,并用统计法估计单故障现象的故障隶属度,得到评判矩阵。通过评判矩阵和级联法值判断原则诊断变压器故障,不但诊断单种变压器故障正确率较高,且诊断多种故障同时存在的变压器故障效果也较理想。

Since there are many mistakes when the faults of transformers are justified by traditional methods such as the IEC three-radio method, many mathematic tools are applied for the faults diagnosis of transformers, for instance, fuzzy identification, ANN （ Artificial neural network）, data mining approach, rough set, grey theory, cluster analysis, expert system and so on. In this paper, due to these shortcomings, a cascaded thresholds technique is proposed accordingly, the weighted fuzzy identification technique based on cascaded thresholds is then achieved. The weighted coefficients matrix are obtained by analyzing the weightiness of different radios between gases for fault diagnosis, for example, the radio between hydrogen and all the fault gases. Faults membership values of single radio are calculated to obtain a judge matrix by using the statistic technique. The faults diagnosis is completed via analyzing judge matrix using cascaded thresholds technique. This technique for diagnosing transformer faults has been validated using field cases, proven to be capable of diagnosing effectively both single-fault transformer and multi- faults transformer.