| 引用本文: | 郑一斌,王慧芳,张磊,姜宽,杨文斌,周才全.基于LightGBM算法的配电网单相接地故障区段定位方法[J].电力自动化设备,2021,41(12): |
| ZHENG Yibin,WANG Huifang,ZHANG Lei,JIANG Kuan,YANG Wenbin,ZHOU Caiquan.Single-phase grounding fault section location in distribution network based on LightGBM algorithm[J].Electric Power Automation Equipment,2021,41(12): |
|
| 摘要: |
| 我国广泛分布的中性点不接地配电网受量测设备配置所限，长期存在故障区段定位困难问题。为此，提出了一种量测量受限条件下的基于轻量级梯度提升机(LightGBM)算法的故障区段定位方法。该方法可在不加装相量测量单元(PMU)、微型PMU、电压互感器等量测装置的情况下，利用常规的相电流有效值准确识别中性点不接地配电网故障区段。通过对目标配电网的量测信息和网架结构分析，采用LightGBM算法离线建立故障区段定位模型，实现在线快速定位故障区段。其中，故障前后各线路稳态电流有效值变化量被选定为故障识别核心特征。此外，针对模型预测结果可能存在的偏差，提出了易错线路标注法和类别概率法辅助识别误判情况，进一步提高了预测结果的可靠性。以修改后的IEEE 123节点系统作为算例，仿真验证了所提方法的可行性和有效性。 |
| 关键词: 故障区段定位 配电网 叠加原理 LightGBM算法 人工智能 |
| DOI：10.16081/j.epae.202108033 |
| 分类号:TM713 |
| 基金项目: |
|
| Single-phase grounding fault section location in distribution network based on LightGBM algorithm |
|
ZHENG Yibin1, WANG Huifang1, ZHANG Lei2, JIANG Kuan1, YANG Wenbin2, ZHOU Caiquan2
|
|
1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;2.POWERCHINA Huadong Engineering Corporation Limited, Hangzhou 311122, China
|
| Abstract: |
| Limited by the configuration of measurement equipment, the fault section location in the isolated neutral distribution network widely distributed in China has been a problem for a long time. Faced with this issue, a fault section location method based on LightGBM(Light Gradient Boosting Machine) algorithm under the condition of limited measurement is proposed. The proposed method can accurately locate the fault section in the isolated neutral distribution network by regular RMS(Root-Mean-Square) values of phase currents without additional installation of PMU(Phasor Measurement Unit),μPMU(micro-Phasor Measurement Unit),voltage transformer and other measuring devices. By analyzing the measurement information and grid structure of the original target distribution network, the LightGBM algorithm is used to establish offline fault section location model to quickly locate the fault section online. Specifically, the variations of the pre-fault and post-fault steady-state RMS values of line currents are taken as the core features. Otherwise, aiming at the possible deviation of model prediction results, the error-prone line labeling and category probability method are used to assist in identifying misjudgment situation which further improves the relia-bility of prediction results. The feasibility and efficiency of the proposed method are verified by using a modified IEEE 123-bus test system. |
| Key words: fault section location distribution network superposition principle LightGBM algorithm artificial intelligence |
