| 引用本文: | 毕贵红,张靖超,赵四洪,等.基于小波散射变换与IRCA-ICA-Res结合的电压源控制型静止同步补偿系统的故障诊断[J].电力系统保护与控制,2025,53(8):144-158.[点击复制] |
| BI Guihong,ZHANG Jingchao,ZHAO Sihong,et al.Fault diagnosis of voltage source controlled static synchronous compensator based on combination of wavelet scattering transform and IRCA-ICA-Res[J].Power System Protection and Control,2025,53(8):144-158[点击复制] |
|
| 摘要: |
| 为了充分利用电压源控制静止同步补偿器(voltage source controlled static synchronous compensator, VSC- STATCOM)IGBT开路故障电流信号中包含的时频信息，以提高IGBT故障诊断和识别的准确性，提出了一种基于小波散射变换(wavelet scattering transform, WST)与改进残差通道注意力(improved residual channel attention, IRCA) 模块、改进坐标注意力(improved coordinate attention, ICA)模块和残差神经网络(residual neural network, Resnet)相结合的新算法—WST-IRCA-ICA-Res。首先，运用Matlab/Simulink平台仿真不同工况下VSC- STATCOM模块22类故障类型，获取故障样本集。其次，利用WST对故障信号进行自动鲁棒的特征提取，构建包含时频信息的特征矩阵。最后，利用IRCA-ICA-Res模型对特征矩阵进行深层次提取、强化和识别。实验结果表明，所提方法具有较强的抗噪性能，能够高精度识别IGBT故障类型。 |
| 关键词: 小波散射变换 注意力模块 神经网络 故障诊断 时频信息 |
| DOI：10.19783/j.cnki.pspc.240926 |
| 投稿时间：2024-03-31修订日期：2024-08-25 |
| 基金项目:国家自然科学基金项目资助(51767012) |
|
| Fault diagnosis of voltage source controlled static synchronous compensator based on combination of wavelet scattering transform and IRCA-ICA-Res |
| BI Guihong,ZHANG Jingchao,ZHAO Sihong,CHEN Shilong,KONG Fanwen,CHEN Dongjing |
| (Kunming University of Science and Technology, School of Electric Power Engineering, Kunming 650500, China) |
| Abstract: |
| To fully utilize the time-frequency information contained in the IGBT open-circuit fault current signals of voltage source controlled static synchronous compensator (VSC-STATCOM) and improve the accuracy of fault diagnosis and identification, a novel WST-RCA-ICA-Res algorithm is proposed. This algorithm combines wavelet scattering transform (WST) with an improved residual channel attention (IRCA) module, and an improved coordinate attention (ICA) module with a residual neural network (Resnet). First, the Matlab/Simulink platform is used to simulate 22 types of faults in the VSC-STATCOM module under different operating conditions to obtain the fault sample set. Then, automatic robust feature extraction of fault signals is carried out using WST to construct a feature matrix containing time-frequency information. Finally, the IRCA-ICA-Res model is used to deeply extract, strengthen, and identify the feature matrix. Experimental results show that the proposed method has strong anti-noise performance and can accurately distinguish different IGBT fault types. |
| Key words: wavelet scattering transform attention module neural network fault diagnosis time-frequency information |
