受消弧线圈、过渡电阻、环境噪声等因素的影响，谐振接地系统故障信号特征微弱，且基于单一判据构成的故障选线方法难以保证选线结果的可靠性。针对上述问题，提出了一种基于动态时间弯曲（DTW）距离算法−Hilbert包络能量与改进K−means聚类算法的谐振接地系统故障选线方法。基于故障线路与健全线路波形相似度差距较大的原理，采用DTW距离算法定量刻画各线路电流序列之间的波形相似程度；为避免单一判据可能存在的选线盲区，基于故障线路与健全线路的能量系数区分度明显的原理，引入Hilbert包络能量衡量暂态零序电流信号中的高频分量幅值；为增强所提选线方法的数据处理能力与效率，采用改进K−means聚类算法对故障特征数据集进行分类处理，将各条线路的故障信息整理为故障数据集，作为改进K−means聚类算法的输入，聚类算法输出各条线路的聚类标签，依据聚类标签判定故障线路。仿真实验结果表明：① 该方法在面对不同过渡电阻、不同故障距离、不同故障初相角、不同线路结构等工况时，均可确保选线结果的准确性；② 相较于传统的K−means聚类算法，改进K−means聚类算法将选线准确率提升了3.4%。现场测试数据表明：该方法具有较强的抗噪声干扰能力，能够在强噪声环境下将保护的耐过渡电阻能力提升至3 000 Ω。

Due to the influence of factors such as arc suppression coils, transition resistance, and environmental noise, the fault signal characteristics in resonant grounding systems are weak. Furthermore, fault line selection methods based on a single criterion often fail to ensure the reliability of the results. To address these issues, this paper proposed a fault line selection method for resonant grounding systems based an dynamic time warping (DTW) distance algorithm-Hilbert envelope energy, and improved K-means clustering algorithm. Based on the principle that the waveform similarity between the faulted and healthy lines differs significantly, the DTW distance algorithm was first employed to quantitatively measure the similarity between current waveforms of each line. To avoid the potential blind spots of a single criterion, Hilbert envelope energy was introduced to measure the high-frequency components in the transient zero-sequence current signals, based on the principle that the energy coefficient distinguishes faulted and healthy lines clearly. Additionally, to enhance the data processing capability and efficiency of the proposed method, the improved K-means clustering algorithm was applied to classify the fault feature dataset. The fault data from each line were organized into a fault dataset, which served as the input to the improved K-means algorithm. The clustering algorithm output the cluster labels for each line, and the fault line was determined based on the cluster labels. Simulation results showed that: ① The method ensured accurate line selection results under various conditions, such as different transition resistances, fault distances, fault initial phase angles, and line structures. ② Compared with the traditional K-means clustering algorithm, the improved K-means algorithm improved the line selection accuracy by 3.4%. Field test data demonstrated the strong noise immunity of the method, improving the protection's tolerance to transition resistance up to 3 000 Ω in a high-noise environment.