基于随机森林二分类器的模块化多电平换流器子模块开路故障检测方法  

A Fault Detection and Location Strategy for Sub-module Open-circuit Fault inModular Multilevel Converters Based on Random Forest Binary Classifier

作　　者：杨贺雅 邢纹硕 陈聪[1] 张伟[2] 李成敏 向鑫[1] 李武华[1] 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)

机构地区：[1]浙江大学,浙江省杭州市310027 [2]内蒙古电力科学研究院,内蒙古自治区呼和浩特010020 [3]洛桑联邦理工学院,瑞士洛桑1015

出　　处：《中国电机工程学报》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)。

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.

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 

分 类 号：TM46[电气工程—电器]