摘要
随着模块化多电平换流器(modularmultilevel converter,MMC)应用领域的日益扩展,其子模块的开路故障引起了更多关注。为了诊断子模块开路故障,该文提出一种基于机器学习(machinelearning,ML)的故障检测和定位策略。根据开路故障特性,文中选择子模块电容器电压作为故障检测的关键指标,然后引入一种从电压数据中提取时域特征的方法,以构造用于有监督学习分类器的样本。在对随机森林的分类器进行样本训练后,检测策略实时电压数据的特征量判断每个子模块的工作状态。所提出的策略可快速准确地定位故障子模块,而无需添加额外的传感器或构建电路的数学模型。最后,通过三相MMC实验平台验证所提出的开路故障检测策略的有效性。
Open-circuit fault of sub-module attracts more attention with the increasing applications of modular multilevel converter(MMC)in high-voltage direct-current(HVDC).This paper proposes a fault detection and location strategy based on machine learning to diagnose the open-circuit fault of sub-module in a modular multilevel converter.According to the open-circuit fault characteristics,the sub-module capacitor voltages are selected as the key indicator for fault detection and location.With feature extraction,the proposed strategy constructs a random forest binary classifier to discriminate the working state of each sub-module.The strategy can locate the faulty sub-module quickly and accurately without extra sensors or mathematical models of the circuit.In order to verify the effectiveness of this strategy,a three-phase MMC prototype is built for data collection.
作者
杨贺雅
邢纹硕
陈聪
张伟
李成敏
向鑫
李武华
YANG Heya;XING Wenshuo;CHEN Cong;ZHANG Wei;LI Chengmin;XIANG Xin;LI Wuhua(Zhejiang University,Hangzhou 310027,Zhejiang Province,China;Inner Mongolia Power Research Institute,Hohhot 010020,Inner Mongolia,China;École Polytechnique Fédérale de Lausanne(EPFL),Lausanne 1015,Switzerland)
出处
《中国电机工程学报》
EI
CSCD
北大核心
2023年第10期3916-3927,共12页
Proceedings of the CSEE
基金
国家重点研发计划国际合作重点专项(2022YFE0101900)
国家自然科学基金项目(52107214,52007166)
内蒙古自治区重大专项(2021D0026)。
关键词
模块化多电平换流器
开路故障检测
时域特征提取
随机森林
二分类器
modular multilevel converter(MMC)
open-circuit fault detection
time-domain feature extraction
random forest
binary classifier
