|
| 基于空间相关性及改进Stacking策略的台区总表故障辨识方法 |
| Fault identification of total meter in substation area based on spatial correlation and improved stacking strategy |
| 投稿时间：2025-11-05 修订日期：2026-01-19 |
| DOI： |
| 中文关键词: 台区总表 故障辨识 空间相关性 改进Stacking 注意力加权 |
| 英文关键词: transformer meter fault identification spatial correlation im-proved stacking attention weighting |
| 基金项目:国家自然科学基金（52307121） |
|
| 摘要点击次数: 0 |
| 全文下载次数: 0 |
| 中文摘要: |
| 配电系统台区总表用于计量与记录台区用电量，其运行状态直接关系到用电管理、线损分析与异常监测等关键环节。针对现有诊断方法过度依赖经验、智能化程度不足的问题，文中提出一种考虑空间相关性的改进Stacking台区总表故障诊断方法。该方法以TimesNet、TimeMixer与InceptionTime作为基学习器，先由注意力加权模块在验证集上学习样本自适应权重，得到一层融合表示；再以LightGBM作为二层元学习器，对加权概率进行非线性组合与误差校正。同时，文中将相邻台区的多变量量测通过主成分分析降维为一维外生特征，与目标台区特征联合输入。文中以某市10kV台区四类故障数据为对象，构建了是否引入相邻台区特征条件下集成模型与单一模型等对比实验，并在多级别噪声条件下开展训练与测试。结果表明，与单一模型相比，考虑空间相关性的改进Stacking策略能有效增强台区总表故障诊断的准确性与抗噪能力。 |
| 英文摘要: |
| This study targets fault diagnosis of distribution transformer area (DTA) master meters, whose health underpins consump-tion management, line-loss analysis, and anomaly monitoring. To overcome heuristic dependence and limited automation in exist-ing methods, it develops a spatial-correlation-aware stacking framework. TimesNet, TimeMixer, and Inception-Time serve as base learners; an attention-weighted module learns sam-ple-adaptive weights on a validation set to form a fused first-layer representation, which a LightGBM me-ta-learner combines nonlinearly for probability calibration and error correc-tion. Multivariate measurements from adja-cent DTAs are com-pressed via principal component analysis into a one-dimensional exogenous feature and concatenated with target-area features. Experiments on four 10-kV DTA fault categories compare the ensemble against individual models, with and without spatial features, under multiple Gaussian-noise levels. Results show consistently higher ac-curacy and reduced degradation under noise, indicating that the proposed spatially informed stacking approach enhances both diagnostic performance and robustness in practical de-ployments. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |