| 引用本文: | 杨兴武,卢愿,王江,徐浩文,孟致丞.基于稀疏自编码器的模块化多电平换流器无监督开路故障诊断策略[J].电力自动化设备,2025,45(7):97-104 |
| YANG Xingwu,LU Yuan,WANG Jiang,XU Haowen,MENG Zhicheng.Unsupervised open-circuit fault diagnosis strategy of MMC based on sparse autoencoder[J].Electric Power Automation Equipment,2025,45(7):97-104 |
|
| 摘要: |
| 基于人工智能的模块化多电平换流器故障诊断方法具有广泛的适用性，但该类方法的模型训练成本大且算法复杂。对开路故障下子模块的电容电流状态进行对比分析，将正常运行时电容电流的理论值、实际值作为原始输入数据；利用稀疏自编码器强大的数据挖掘能力来设置多分类子模块的故障诊断阈值，从而直接实现故障的准确定位和类型判断。所提策略采用无监督学习的方式进行模型训练，无须进行各类故障样本和数据处理即可快速准确地实现不同场景的多管故障诊断。相较于现有基于人工智能的诊断策略，所提策略的鲁棒性更强，诊断速度更快，计算成本更低，且规避了手动阈值选取复杂的问题。仿真和实验结果验证了所提策略的有效性。 |
| 关键词: 模块化多电平换流器 人工智能 无监督学习 稀疏自编码器 故障诊断 |
| DOI：10.16081/j.epae.202501007 |
| 分类号:TM46 |
| 基金项目:上海市科技计划项目(23010501200) |
|
| Unsupervised open-circuit fault diagnosis strategy of MMC based on sparse autoencoder |
|
YANG Xingwu1, LU Yuan1, WANG Jiang2, XU Haowen1, MENG Zhicheng1
|
|
1.College of Electrical Engineering, Shanghai University of Electric Power, Shanghai 200090, China;2.Baoji Power Supply Company, State Grid Shaanxi Electric Power Co.,Ltd.,Baoji 721000, China
|
| Abstract: |
| The fault diagnosis methods of modular multilevel converter(MMC) based on artificial intelligence have wide applicability, but the model training cost of these methods is high and the algorithm is complex. The capacitance current states of sub-modules with open-circuit fault are compared and analyzed, and the theoretical and actual values of capacitance current during normal operation are taken as the original input data. The powerful data mining ability of sparse autoencoder is utilized to set the diagnosis threshold values of multiple types of sub-module faults, thereby directly achieving the accurate location and type determination of faults. The proposed strategy adopts an unsupervised learning approach for model training, and can quickly and accurately achieve multi-switching tube fault diagnosis under different scenarios without dealing with various fault samples and data. Compared with the existing artificial intelligence-based diagnosis strategies, the proposed strategy has stronger robustness, faster diagnosis speed and lower computing cost, and avoids the complex problem of manual threshold value selection. Simulative and experimental results verify the effectiveness of the proposed strategy. |
| Key words: modular multilevel converter artificial intelligence unsupervised learning sparse autoencoder fault diagnosis |
