雷电冲击下换流变压器绕组故障模拟与诊断方法

Fault Simulations and Diagnosis Methods of Converter Transformer Windings Under Lightning Impulse Voltages

现有的雷电冲击下换流变压器绕组故障诊断方法灵敏度较低,且难以判断故障类型;绕组在雷电冲击下发生故障的试验数据较少,难以深入分析故障特征。针对上述问题,设计并制作了具有典型换流变压器绕组结构、比例约为1/5的缩比模型,并模拟不同故障类型进行试验研究。分析故障情况下的试验数据,利用故障情况与正常情况传递函数不同频率段的相关系数作为特征参数;提出了基于学习矢量量化(learning vector quantization,LVQ)网络的变压器绕组短路故障诊断方法,利用特征参数对LVQ网络进行训练,并对训练后的网络进行测试。结果表明:通过特定频率段相关系数的对比可以区分不同故障;LVQ网络故障诊断测试准确率达到98%,验证了诊断方法的有效性。

The existing fault diagnosis methods of converter transformer windings under lightning impulse voltages have low sensitivity. It is difficult to diagnose fault types. The experimental data of the converter transformer under fault conditions are insufficient. It is difficult to study on the fault characteristics deeply. In view of the above problems, the reduced-scale model with a typical structure of converter transformer windings was designed and made with the scale of about 1/5. The lightning impulse tests were carried out under different fault conditions. The test data of different fault conditions were analyzed. The correlation coefficients for the transfer function of the fault condition and normal condition was calculated in different frequency bands and chosen as characteristic parameters. A fault diagnosis method was proposed based on the learning vector quantization network. The network was trained with characteristic parameters. Then the testing samples were used for testing. The results show that the correlation coefficients in specified frequency bands can reflect fault characteristics. The diagnostic accuracy of the testing samples is 98.2%. It verifies the effectiveness of the fault diagnosis method.