基于广义神经网络与模糊聚类的变压器故障诊断

Transformer Fault Diagnosis Based on GRNN and FCM

鉴于IEC三比值法在变压器故障诊断中,存在编码缺失和编码边界过于绝对等缺陷,提出了基于广义回归神经网络(GRNN)和模糊C-均值聚类算法(FCM)的变压器故障诊断方法,建立了GRNN-FCM联合变压器故障诊断模型。选取变压器油中5种特征气体体积分数及其三比值编码作为输入特征向量,利用GRNN模型对样本故障进行初步判断(正常、过热、放电、放电兼过热),再采用模糊C-均值聚类算法对样本故障作进一步判断,最终得到具体的故障类型。将该模型与其他几种故障诊断方法进行对比分析,仿真实验结果表明,GRNN-FCM联合变压器故障诊断模型输出值与实际值具有较好一致性且准确度更高,验证了该模型的可行性及实用性。

In view of the problem that IEC three ratio method has the disadvantage of absolute boundary and lack of codes, a novel transformer fault diagnosis method based on generalized regression neural network（GRNN）and Fuzzy C-means（FCM）clustering algorithm is proposed and a GRNN-FCM combined transformer fault diagnosis model is constructed. Five volume fractions of the gases in oil and its three-ratio codes are chosen to be the inputs of the combined model. This model uses GRNN model to judge the preliminary fault type（normal, overheat,discharge,discharge overheat）,then Fuzzy C-means clustering algorithm is adopted to achieve the transformer fault diagnosis. After comparing the combined model to other two diagnosis methods, the simulation results indicate that this combined model has accordant outputs with the measured values and provides higher accuracy,so the feasibility and effectiveness of the model presented are verified.