| 引用本文: | 金立军,张志伟,郑含博,等.基于物理信息网络的高压断路器分合闸电流分析及故障诊断[J].电力系统保护与控制,2026,54(10):162-172. |
| JIN Lijun,ZHANG Zhiwei,ZHENG Hanbo,et al.Analysis of opening/closing coil current and fault diagnosis of high-voltage circuit breakers based on physics-informed neural networks[J].Power System Protection and Control,2026,54(10):162-172 |
|
| 摘要: |
| 针对传统的高压断路器故障诊断模型存在机理缺失与数据依赖等问题，提出一种基于多输出物理信息网络的分合闸线圈电流动态特性建模及故障诊断方法。首先，依据分合闸线圈动作机理搭建电气 - 机械物理场耦合的数学模型，建立物理先验约束函数。然后，考虑线圈电流与动铁芯运动存在强耦合关系，将微分方程以残差损失形式嵌入相应的分支网络。并引入自适应权重调整策略平衡各项梯度流，实现分合闸电流动态特性快速准确地正向推演与分析。最后，提出结合迁移学习思想的故障诊断方法，将先验约束模型向物理实体迁移。试验结果表明，所提方法具有较高的物理一致性与求解精度，即便在较小实测数据规模下故障诊断准确率仍达到 96.86%。 |
| 关键词: 高压断路器 分合闸线圈 物理信息神经网络 迁移学习 故障诊断 |
| DOI：10.19783/j.cnki.pspc.251210 |
| 分类号: |
| 基金项目:国家自然科学基金项目资助 (52277157) |
|
| Analysis of opening/closing coil current and fault diagnosis of high-voltage circuit breakers based on physics-informed neural networks |
|
JIN Lijun1, ZHANG Zhiwei1, ZHENG Hanbo2, LI Jinheng2, WANG Jinyu1, YAO Siyu1
|
|
1. School of Electronics and Information Engineering, Tongji University, Shanghai 201804, China; 2. School of Electrical Engineering, Guangxi University, Nanning 530004, China
|
| Abstract: |
| To address the issues of mechanism deficiency and heavy data dependency in traditional high-voltage circuit breaker fault diagnosis models, a method for dynamic characteristic modeling and fault diagnosis of opening/closing coil currents based on a multiple-output physics-informed neural network is proposed. First, an electromechanical coupled mathematical model is constructed based on the operating mechanism of the opening/closing coils, and corresponding physical prior constraint functions are established. Then, considering the strong coupling relationship between coil current and the motion of the moving iron core, the differential equations are embedded into the corresponding branch networks in the form of residual loss. An adaptive weight adjustment strategy is introduced to balance the gradient flows among different loss terms, enabling fast and accurate forward inference and analysis of the dynamic characteristics of opening/closing currents. Finally, a fault diagnosis method integrated with transfer learning is proposed, allowing the prior-constrained model to be transferred to physical systems. Experimental results demonstrate that the proposed method achieves high physical consistency and solution accuracy, attaining a fault diagnosis accuracy of up to 96.86% even under limited measured data. |
| Key words: high-voltage circuit breaker opening/closing coil physics-informed neural network transfer learning fault diagnosis |
