电工技术学报  2023, Vol. 38 Issue (10): 2744-2756    DOI: 10.19595/j.cnki.1000-6753.tces.220777
电力电子 |
基于多元高斯分布异常检测模型的MMC子模块开路故障诊断方法
杨贺雅1, 邢纹硕1, 向鑫1, 张伟2, 胡宏彬2
1.浙江大学电气工程学院 杭州 310027;
2.内蒙古电力科学研究院 呼和浩特 010020
A Sub-Module Open-Circuit Fault Detection and Location Strategy for Modular Multilevel Converters Based on Multivariate Gaussian Distribution
Yang Heya1, Xing Wenshuo1, Xiang Xin1, Zhang Wei2, Hu Hongbin2
1. College of Electrical Engineering Zhejiang University Hangzhou 310027 China;
2. Inner Mongolia Power Research Institute Hohhot 010020 China
全文: PDF (3605 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 模块化多电平换流器(MMC)的子模块开路故障隐蔽性强,可以及时有效地识别故障位置,防止故障蔓延影响其他组件,提高换流系统的可靠性。为了诊断子模块开路故障,该文提出一种基于多元高斯分布异常检测模型的故障检测方法。首先,分析子模块开路故障特性,选择子模块电容电压作为故障检测的关键指标,并提取子模块电容电压的12个特征量组成特征向量用于故障诊断。然后,根据多元高斯分布特性,提出基于多元高斯分布的异常检测模型构建方法,并基于此模型提出子模块开路故障诊断策略,可在故障样本偏少致使样本不均衡时,实现高准确率、高效率的故障诊断。最后,通过仿真和实验验证了所提出的MMC子模块开路故障诊断方法的有效性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
杨贺雅
邢纹硕
向鑫
张伟
胡宏彬
关键词 模块化多电平换流器开路故障检测多元高斯分布    
Abstract:Modular multilevel converter (MMC) is drawing increasing attention in high-voltage direct- current (HVDC) transmission. Due to the numerous cascaded sub-modules (SMs), safety is one of the most important challenges for MMC. The open-circuit faults for SMs may distort the output voltage and current of MMC and even destroy the system if the faults are not handled in time. Therefore, this paper will concentrate on SM open-circuit faults. In general, the research for fault detection and location (FDL) of SM open-circuit is mainly divided into three basic approaches: (1) sensor-based, (2) model-based, and (3) machine learning (ML)-based. Sensor-based methods can realize FDL quickly but need extra hardware costs, which may also add potential points of fault. Model-based methods are relatively sensitive to uncertain system parameters and environmental noise because of the requirement for an accurate mechanism model of the circuit. This paper proposes a diagnostic strategy based on machine learning (ML) to detect and locate the sub-module open-circuit fault.
Firstly, the operation principle of MMC and SMs are analyzed to find that the capacitor voltages of faulty SMs will be different from the normal capacitance voltage. SM capacitor voltages can be regarded as the indicator for fault detection and location. The sliding window divides the long original data into multiple short data fragments, and 12 features of SM capacitor voltages are extracted as the feature vectors for the anomaly detection model. Secondly, the open circuit fault diagnosis problem can be transformed into a multivariate Gaussian anomaly detection problem according to the feature extraction of the capacitor voltage of SMs. The model based on multivariate Gaussian distribution is demonstrated to make predictions for anomaly detection and trace back the predicted faulty SM. Then, the fault detection and location method based on multivariate Gaussian distribution is presented to locate the faulty SM. If the probability density of the vector to be diagnosed is greater than the threshold value, it can be determined that the SM is normal. If the probability density is less than the threshold value, the SM is in a fault state. When the open circuit fault is diagnosed for N consecutive cycles, the SM is judged as faulty. Finally, simulation and experiment verify the effectiveness of the proposed FDL method.
In summary, an SM open-circuit FDL method based on multivariate Gaussian distribution is proposed in this paper. After the analysis of the fault characteristics, the voltages of SM are chosen as the indicators of FDL. Sliding window and feature extraction are used to construct the dataset for the ML model. The model based on multivariate Gaussian distribution is trained for anomaly detection. After the fault is detected, it can be located by tracing the corresponding SM number. This method can get high generalization accuracy without extra sensors or an accurate mathematical model of MMC. According to the simulation and experiment results, the effectiveness of the proposed method can be proven by the different evaluation indexes.
Key wordsModular multilevel converter    open-circuit fault detection    multivariate Gaussian distribution   
收稿日期: 2022-05-10     
PACS: TM46  
基金资助:国家自然科学基金资助项目(52007166, 52107214)
通讯作者: 向 鑫 男,1990年生,博士,研究员,研究方向为新能源发电并网和柔性直流输配电技术。E-mail: xiangxin@zju.edu.cn   
作者简介: 杨贺雅 女,1991年生,博士,特聘副研究员,研究方向为大功率变流与控制技术。E-mail: yangheya@zju.edu.cn
引用本文:   
杨贺雅, 邢纹硕, 向鑫, 张伟, 胡宏彬. 基于多元高斯分布异常检测模型的MMC子模块开路故障诊断方法[J]. 电工技术学报, 2023, 38(10): 2744-2756. Yang Heya, Xing Wenshuo, Xiang Xin, Zhang Wei, Hu Hongbin. A Sub-Module Open-Circuit Fault Detection and Location Strategy for Modular Multilevel Converters Based on Multivariate Gaussian Distribution. Transactions of China Electrotechnical Society, 2023, 38(10): 2744-2756.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.220777          https://dgjsxb.ces-transaction.com/CN/Y2023/V38/I10/2744