基于多任务学习的新型电力系统故障诊断方法  

New power system fault diagnosis method based on multi⁃task learning

作　　者：高陆军 陈洁[1] 王新雷 田雪沁 德格吉日夫 GAO Lujun;CHEN Jie;WANG Xinlei;TIAN Xueqin;Degejirifu(School of Electrical Engineering,Xinjiang University,Urumqi 830017,China;State Grid Economic and Technological Research Institute Co.,Ltd.,Beijing 102209,China)

机构地区：[1]新疆大学电气工程学院,新疆乌鲁木齐830017 [2]国网经济技术研究院有限公司,北京102209

出　　处：《现代电子技术》2023年第15期155-160,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(61963034);新疆维吾尔自治区自然科学基金资助项目(2022D01C366)

摘　　要：当前对于电力系统故障诊断的研究大都基于传统电力系统,而且对于不同诊断任务采用不同的诊断模型,这在实际应用中不仅造成了设备资源的浪费,也忽略了各诊断任务之间的关联。为此,运用多任务联合训练方式挖掘不同诊断任务之间的关联信息,建立一种多任务学习的新型电力系统故障诊断模型。首先通过底层共享深度置信网络(DBN)层进行特征提取;然后通过顶层分类层输出诊断结果;最后实验结果表明,所提模型不仅能够实现新型电力系统的故障类型和故障选线诊断任务,而且具有较高的效率以及精度。除此之外,还与同结构的单任务学习诊断模型进行对比,结果表明,多任务学习模型诊断效果更好。The current research on power system fault diagnosis is mostly based on the traditional power system,and different diagnostic models are adopted for different diagnostic tasks,which not only cause the waste of equipment resources,but also neglect the correlation between each diagnostic task in practical applications.Therefore,the multi⁃task joint training method is used to mine the correlation information among different diagnostic tasks,so as to establish a new power system fault diagnosis model based on multi⁃task learning.The underlying shared deep belief network(DBN)layer is used to extract the features,and then the diagnostic results are output by the top⁃level classification layer.The experimental results show that the proposed model can not only realize the diagnosis task of fault type and fault line identification,but also have high efficiency and accuracy.In addition,in comparison with the single⁃task learning diagnosis model with the same structure,the model based on multi⁃task learning has better diagnosis effect.

关 键 词：新型电力系统 深度置信网络 多任务学习 故障诊断 单任务学习 特征提取 联合训练 

分 类 号：TN99-34[电子电信—信号与信息处理] TM726[电子电信—信息与通信工程]