基于格拉姆角场与ResNet的输电线路故障辨识方法

针对如何利用实际故障录波数据，提取和放大故障特征差异，开展故障类型与故障原因辨识的问题，提出了基于格拉姆角场与迁移学习-ResNet的输电线路故障辨识方法。首先，统计分析了输电线路故障类型和故障原因的分布特征，用于指导构建适用于类不平衡问题的故障分类器。然后，利用格拉姆角场变换将采集得到的故障电压、电流时序信号转化为格拉姆角场图像，放大故障特征差异，作为故障分类器的输入。进一步，将生成的图像集输入搭建好的故障分类器进行网络训练和测试，输出输电线路故障类型和故障原因。最后，完全采用真实故障录波数据开展了算例分析。结果表明：所提方法对故障类型的辨识准确率达到了97.51%，对故障原因的辨识准确率达到了94.23%。并且将训练的故障辨识网络迁移至其他地区时，仍然具有较好的故障辨识效果和泛化性能。所提方法为基于暂态波形数据驱动的故障辨识提供了新方法，可以用于实际电网的输电线路故障辨识。

Transmission line fault identification method based on Gramian angular field and ResNet

There is a problem of how to use actual fault recorded data to extract and amplify fault feature differences, and carry out fault type and cause identification. Thus a fault identification method for transmission lines based on Gramian angular field (GAF) and transfer learning-ResNet is proposed. First, the distribution characteristics of fault type and cause on transmission lines are analyzed. These are used to guide the construction of fault classifiers suitable for a class imbalance problem. Second, the collected fault voltage and current time sequence signals are converted into GAF images by GAF transform, so that the fault feature differences are amplified as the input of the fault classifier. The generated GAF image set is then fed into an established fault classifier for network training and testing, and the type and cause of transmission line faults are output. Finally, an example analysis using real fault recorded data shows that the proposed method has achieved 97.51% accuracy for fault type identification and 94.23% accuracy for fault cause identification; the trained fault identification network still achieves good fault identification and generalization performance when transferred to other regions. The proposed method provides a novel method for fault identification based on transient waveform data. It can be used for transmission line fault identification in practical power grids.