面向电力设备异常检测的深度自编码支持向量数据描述模型研究  

Research on Deep Auto Encoding Support Vector Data Description Model for Power Equipment Anomaly Detection

作　　者：耿波[1] 潘曙辉 董晓旭 GENG Bo;PAN Shuhui;DONG Xiaoxu(Research Institute of Nuclear Power Operation,Wuhan 430223,China;School of Computer Science and Technology,Huazhong University of Science and Technology,Wuhan 430074,China;School of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063210,China)

机构地区：[1]核动力运行研究所,湖北武汉430223 [2]华中科技大学计算机科学与技术学院,湖北武汉430074 [3]华北理工大学冶金与能源学院,河北唐山063210

出　　处：《湖南电力》2024年第1期119-127,共9页Hunan Electric Power

基　　金：国家自然科学基金青年基金资助项目(52004094);河北省自然科学基金项目(E2021209037);中央引导地方科技发展基金项目(236Z1017G)。

摘　　要：针对深度自编码支持向量数据描述模型对电力设备部分异常区分能力不足的问题,提出自监督混合专家增强的深度自编码支持向量数据描述模型,构造多种自监督变换数据集模拟潜在未知异常,引入自监督分类和掩码重构任务以学习更具区分性的表示。此外,将编码器部分改造为混合专家模型结构,将数据分配给不同专家子模块进行专业化的学习,使异常决策边界更清晰。在4个公开数据集和3个电厂设备数据集上的实验结果证实了自监督学习和混合专家模型的有效性。In response to the problem of insufficient ability of deep auto encoding support vector data description model to distinguish some anomalies of power equipment,this paper proposes deep auto encoding support vector data description model with self-supervised and mixture-of-experts.The multiple self-supervised transformation datasets is constructed to simulate potential unknown anomalies,and the self-supervised classification and mask reconstruction tasks is introduced to learn more discriminative representations.In addition,the encoder is transformed into a mixture-of-experts structure,and the data is allocated to different expert sub-modules for professional learning,making the abnormal decision boundary clearer.Experimental results from four public data sets and three power plant equipment data sets prove the effectiveness of self-supervised learning and mixture-of-experts.

关 键 词：异常检测 深度自编码支持向量数据描述 自监督学习 混合专家模型 

分 类 号：TM762[电气工程—电力系统及自动化]