| 引用本文: | 王江波,黑晓捷,邱鹏玉,胡旭峰,罗晶晶,何啸扬.基于XGBoost和泛化特征优选的小电流接地故障方向判别方法[J].电力自动化设备,2025,45(7):71-79 |
| WANG Jiangbo,HEI Xiaojie,QIU Pengyu,HU Xufeng,LUO Jingjing,HE Xiaoyang.Small current grounding fault direction discrimination method based on XGBoost and generalization feature optimization[J].Electric Power Automation Equipment,2025,45(7):71-79 |
|
| 摘要: |
| 考虑配电网的复杂多变性，提出了一种基于极限梯度提升(XGBoost)和泛化特征优选的故障方向判别方法，以数量最少且能适应不同场景变化的最优泛化特征作为输入，提升故障方向判别模型的准确率和泛化能力。考虑到实际应用中电流信号的易获取性，以归一化三相暂态电流突变量波形级联构造特征波形，提取6类26个特征量构建候选特征集；利用多场景精细化仿真数据集拟合各特征类间概率分布，基于Hellinger距离稳健引导和互补排序，以模型准确率达到稳定的最少特征数为准则确定最优泛化特征子集。与不同分类算法的对比分析结果验证了所提方法有效性，特征优选后仅需输入4个特征即可达到较高的判别准确率。同时，经高阻接地故障识别能力分析、噪声模拟测试和实测故障录波数据验证，所提方法的准确率可达99 % 以上，表明所提方法具有较好的泛化性和鲁棒性。 |
| 关键词: 小电流接地故障 故障方向 故障特征 故障分析 XGBoost模型 |
| DOI：10.16081/j.epae.202506001 |
| 分类号:TM713 |
| 基金项目:国家自然科学基金智能电网联合基金资助项目(U2166208) |
|
| Small current grounding fault direction discrimination method based on XGBoost and generalization feature optimization |
|
WANG Jiangbo1, HEI Xiaojie1, QIU Pengyu2, HU Xufeng1, LUO Jingjing1, HE Xiaoyang1
|
|
1.College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China;2.Jinchuan Group Co.,Ltd.,Jinchang 737100, China
|
| Abstract: |
| Considering the complexity and variability of the distribution network, a fault direction discrimination method based on extreme gradient Boosting(XGBoost) and generalization feature optimization is proposed. The optimal generalization features with the least number and the adaptability to different scenarios are used as input to improve the accuracy and generalization ability of the fault direction discrimination model. Considering the easy acquisition of signals in practical applications, the feature waveform is construc-ted by cascading the normalized three-phase transient current mutation waveform, and 26 features of 6 categories are extracted to construct the candidate feature set. The multi-scene refined simulation data set is used to fit the probability distribution between feature classes. Based on Hellinger distance robust guidance and complementary sorting, the optimal generalization feature subset is determined by the minimum number of features with stable model accuracy. The effectiveness of the method is verified by comparing with diffe-rent classification algorithms. After feature selection, only four features are input to achieve higher discrimination accuracy. At the same time, the accuracy rate can reach more than 99 % through the analysis of high resistance grounding fault identification ability, noise simulation and measured fault recording data test, which shows that the proposed method has good generalization and robustness. |
| Key words: small current grounding fault fault direction fault feature fault analysis XGBoost model |
