基于CBAM-ResNet和多域特征融合的配电网故障选线方法

Distribution network fault line selection method based on multi-domain feature fusion and improved two-branch residual network

传统配电网故障选线模型大多采用零序电流时频特征作为选线判据,单一信息域特征无法全面反映故障信息差异、适用范围存在局限性。为了提高模型复杂工况下选线准确率,提出一种基于卷积注意力机制优化双分支残差网络(CBAM-ResNet)和多域特征融合的配电网故障选线方法。首先,利用变分模态分解-希尔伯特变换和格拉姆角场将采集的零序电流信号分别映射为二维时频域和空间域图像,构建能够全面的反映故障信息的多域图像训练集;其次,通过CBAM-ResNet网络深层次挖掘并融合多域特征信息,卷积注意力机制能对多域特征的重要性进行区分,加快网络训练速度,提高分类准确性;最后,将融合特征输入全连接层实现对配电网故障线路的选取。仿真结果表明,该方法相比传统选线方法具有更高的选线精度和噪声鲁棒性。

The traditional fault line selection model of distribution network mostly adopts the time-frequency characteristics of zero-sequence current as the line selection criterion.The single information domain characteristics can not fully reflect the difference of fault information,and the application scope is limited.In order to improve the accuracy of line selection under complex working conditions of the model,a fault line selection method for distribution network based on convolutional attention mechanism optimized two-branch residual network(CBAM-ResNet)and multi-domain feature fusion is proposed.Firstly,the collected zero-sequence current signals are mapped into two-dimensional timefrequency domain and spatial domain images by using variational mode decomposition-Hilbert transform and Gram angle field,respectively,to construct a multi-domain image training set that can fully reflect the fault information.Secondly,the CBAM-ResNet network is used to deeply mine and integrate multi-domain feature information.The convolutional attention mechanism can distinguish the importance of multi-domain features,accelerate network training speed and improve classification accuracy.Finally,the fusion feature is input into the full connection layer to select the fault line of the distribution network.The simulation results show that the proposed method has higher line selection accuracy and noise robustness than the traditional line selection method.