基于卷积神经网络的变压器有载分接开关故障识别

为进一步提高变压器有载分接开关(OLTC)故障识别的精度,从OLTC切换过程中振动信号递归图的纹理特征出发,提出了一种基于卷积神经网络(CNN)的变压器OLTC故障识别方法。首先根据OLTC振动信号的相空间分布,基于相点距离映射构建了OLTC振动信号的距离映射递归图(DMRP),然后通过合理选取CNN的网络层数、卷积核尺寸等结构超参数和对卷积核进行降维处理,提出了基于CNN的OLTC故障识别模型。对某CM型OLTC正常与典型故障下振动信号的计算结果表明,DMRP能自适应地对振动信号的相空间相点分布进行描述,所提出的识别模型对OLTC的典型故障均具有良好的识别性能,尤其在轻微故障的识别上相比于现有方法准确率提升了至少10%。

Fault Recognition of On-load Tap-changer in Power Transformer Based on Convolutional Neural Network

To further improve the fault recognition accuracy of on-load tap-changer (OLTC) of power transformer, this paper presents a fault recognition method of OLTC based on convolutional neural network (CNN) considering the texture features of recurrence plot for vibration signal during the switching process of OLTC. The distance mapping recurrence plot (DMRP) is constructed by using the phase point distance mapping according to the phase space distribution of vibration signals for OLTC. Then a CNN based OLTC fault recognition model is proposed by reasonably selecting the structure hyper-parameters of CNN, including the number of network layers and convolution kernel size, and reducing the convolution kernel dimensionality. The calculation results of vibration signals from a CM type OLTC under normal and typical mechanical fault conditions show that the DMRP can adaptively describe the point distribution of the vibration signals in phase space. The recognition model has excellent recognition performance on OLTC typical faults. Especially, the recognition rate of slight fault is improved by at least 10% compared with the existing methods.