基于1DCNN-BiLSTM的配网单相接地故障选线方法  

Line Selection Method for Single-Phase Grounding Fault in Distribution Network Based on 1DCNN-BiLSTM

作　　者：聂祥论 张垚 申江兰 毛业涛 聂晶 李义 饶雪梅 

机构地区：[1]贵州电网有限责任公司毕节供电局,贵州 毕节

出　　处：《智能电网（汉斯）》2023年第5期83-91,共9页Smart Grid

摘　　要：针对单相接地故障特征信息难以充分挖掘导致故障选线精度不高、鲁棒性差的问题,提出了一种基于一维卷积神经网络(one-dimensional convolutional neural network, 1DCNN)-双向长短期记忆网络(bi-directional long short-term memory, BiLSTM)的配网单相接地故障选线方法。首先,利用序列特征融合方法对同一工况下各线路的暂态零序电流,获取序列融合特征向量。其次,采用1DCNN提取序列融合特征向量的局部特征,BiLSTM进一步从局部特征中学习上下文依赖关系,最后通过SoftMax层实现故障选线。仿真表明,所提方法的选线精度为100%。与现有方法相比,所提方法同时兼顾选线精度高和鲁棒性强的优点。Aiming at the problem that the single-phase grounding fault feature information is difficult to be fully mined leading to low line selection accuracy and poor robustness, a single-phase grounding line selection method based on a one-dimensional convolutional neural network-bi-directional long and short-term memory for distribution networks is proposed. First, the transient zero-sequence currents of each line under the same operating condition are spliced using the serial feature fusion method to obtain the sequence fusion feature vector. Second, 1DCNN is used to extract the local features of sequence fusion feature vectors, BiLSTM further learns the contextual dependencies from the local features, and finally line selection is implemented through SoftMax layer. Simulation shows that the proposed method has line selection accuracy of 100%. Compared with the existing methods, the proposed method combines the advantages of high line selection accuracy and ro-bustness at the same time.

关 键 词：序列特征融合 故障选线 一维卷积神经网络 双向长短期记忆网络 特征提取 上下文依赖关系 

分 类 号：TM8[电气工程—高电压与绝缘技术]