多端柔性直流输电线路单极接地故障定位方法

Single-pole Ground Fault Location Method for Multi-terminal Flexible DC Transmission Line

精准可靠的输电线路故障定位方法对于维持多端柔性直流系统稳定运行至关重要。为解决过渡电阻、行波色散对线路测距的干扰,有效提高输电线路故障定位精度。以先定区段再定位的思想,提出一种采用小波包奇异熵和一维卷积神经网络(Convolutional Neural Networks,CNN)的多端柔性直流输电线路单极接地故障定位方法。在接地故障发生时,提取不同区段线模电压组成特征向量,结合1D-CNN分类模型完成区段识别。故障区段确定后,利用小波包奇异熵提取故障区段双端线模电压的深层故障特征,并基于特征提取结果建立1D-CNN回归模型进行故障定位。为避免模型训练时陷入局部最优,采用麻雀搜索算法(Sparrow Search Algorithm,SSA)对1D-CNN模型进行参数寻优。利用PSCAD/EMTDC建立±500KV四端柔性直流仿真系统模型,进行了多种工况的单极接地故障仿真与定位性能测试。仿真结果表明,所提定位方法具有良好的耐过度能力,在50kHz的采样频率下定位误差保持在0.22km以内。

Accurate and reliable fault location method of transmission line is very important to maintain the stable operation of multi-terminal flexible DC system.In order to solve the interference of transition resistance and traveling wave dispersion on line ranging accuracy,the transmission line fault location accuracy can be improved effectively.In this paper,a method for single-pole ground fault location of multi-terminal flexible HVDC lines is proposed by using wavelet packet singular entropy and one-dimensional Convolutional Neural Networks(CNN).When the ground fault occurs,the feature vector composed of different section line mode voltages is extracted,and the section identification is completed by combining the 1D-CNN classification model.After the fault region was determined,the wavelet packet singular entropy was used to extract the deep fault features of the double-terminal mode voltage in the region,and the 1D-CNN regression model was established based on the feature extraction results for fault location.In order to avoid falling into local optimum during model training,Sparrow Search Algorithm(SSA)was used to optimize parameters of 1D-CNN model.PSCAD/EMTDC was used to establish a±500kV four-terminal flexible DC simulation system model,and unipolar grounding fault simulation and positioning effect test were carried out in various working conditions.The simulation results show that the proposed positioning method has good transition resistance,and the positioning error is kept within 0.22km at the sampling frequency of 50kHz.