判别式
依赖关系(UML)
相似性(几何)
计算机科学
人工智能
电阻抗
模式识别(心理学)
噪音(视频)
断层(地质)
故障检测与隔离
鉴定(生物学)
接地
子网
钥匙(锁)
高阻抗
分布(数学)
特征学习
机器学习
工程类
异常检测
航程(航空)
功率(物理)
噪声测量
监督学习
深度学习
数据挖掘
图像(数学)
可见的
作者
Yongjia Huang,Zeyin Zheng,Guo Moufa,Wei Gao
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-11-26
卷期号:: 168-172
标识
DOI:10.1109/psre67836.2025.11397984
摘要
High impedance ground faults (HIGF) in distribution networks present considerable challenges for detection. They typically exhibit minimal observable signatures and display a high degree of randomness. Conventional techniques for HIGF detection often depend on extensive labeled datasets to establish threshold values. This dependency not only incurs significant labor costs but also struggles to extract deeper latent information from fault data, potentially leading to misclassification. To address these limitations, this research introduces a self-supervised contrastive learning framework for HIGF detection in distribution networks. The proposed approach learns generalizable features directly from unlabeled zero-sequence current (ZSC) samples by maximizing similarity between augmentations of the same sample, thereby minimizing reliance on labeled data. Subsequent fine-tuning with a minimal amount of labeled ZSC data enables accurate identification of high impedance grounding faults. Simulations validate that the model effectively learns discriminative features from ZSC waveforms. Notably, it achieves performance comparable to fully supervised models while utilizing only 1 % of their labeled training data. Additionally, it exhibits superior noise resistance in systems with signal-tonoise ratios of$\mathbf{3 5 d B} \boldsymbol{,} \mathbf{3 0 d B}$, and$\mathbf{2 5 d B}$.
科研通智能强力驱动
Strongly Powered by AbleSci AI