基于空间域图像生成和混合卷积神经网络的配电网故障选线方法

Fault Line Selection for Distribution Network Based on Spatial Domain Image Generation and Hybrid Convolutional Neural Network

传统的配电网故障选线方法大多基于一维零序电流序列构建故障诊断模型,单一的诊断模型往往限制了故障特征的深层挖掘。为了提高故障选线的准确率,提出一种基于空间域图像和混合卷积神经网络的配电网故障选线方法。首先,利用优化的降噪光滑模型对零序电流信号进行降噪处理,减少外界环境的电磁干扰。其次,利用对称希尔伯特变换将一维时域信号转成二维空间域图像,图像的颜色、形状和纹理特征能够充分反映当前系统的运行状态。最后,将一维时域信号和二维空间域图像同步作为混合卷积神经网络的输入,充分挖掘系统的故障特征,利用Sigmoid函数实现故障选线。在辐射状配电网、IEEE-13节点模型、IEEE-34节点、StarSim仿真平台上模型上进行了实验验证。实验结果表明,该选线方法可以有效克服传统方法过度依赖主观特征选择、抗噪性能差等问题,能够在高阻接地、采样时间不同步、两点接地故障等极端情况下可靠地筛选出故障线路。

The traditional distribution network fault line selection methods mostly build fault diagnosis models based on the one-dimensional zero-sequence current signals,while the single diagnosis model often limits the deep mining of the fault features.In order to improve the accuracy of fault line selections,and combing the spatial domain image and hybrid convolutional neural network,a fault line selection for the distribution network is proposed.First,the optimal smooth denoising model is used to depress the noise for the zero-sequence current signals,realizing the reduction of the electromagnetic interference from the external environment.Then,the 1D time-domain signal is transformed into a 2D spatial domain image using the symmetrized Hilbert transform pattern.The color,shape,and texture features of the image can then fully reflect the current system operation status.Finally,the 1D signal and the 2D image are synchronized as the inputs of the hybrid convolutional neural network for fully excavating fault characteristics of the system.The Sigmoid function is used to achieve the fault line selection.Experimental verifications are carried out on the different topologies.Even in the extreme cases,such as the noise interference,the high-impedance grounding,asynchronous sampling,and two-point grounding,the proposed method is still able to accurately select the fault lines.