基于深度置信网络的电能质量扰动事件分类  

Deep Belief Networks Based Classification of Power Quality Disturbance Events

作　　者：王玥 肖斐[1] 艾芊[1] 张宇帆 李昭昱 WANG Yue;XIAO Fei;AI Qian;ZHANG Yufan;LI Zhaoyu(Department of Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电气工程系,上海200240

出　　处：《供用电》2019年第1期40-45,53,共7页Distribution & Utilization

基　　金：国家自然科学基金项目(51577115)~~

摘　　要：为满足电能质量扰动准确分类的需求,提出了一种基于极大重叠离散小波变换(MODWT)和深度置信网络(DBN)的电能质量扰动分类方法。首先利用MODWT提出一种可靠的电能质量暂态事件检测算法,该算法无需设定检测阈值,可准确获取暂态事件的起止时刻。接着提取暂态事件的电压谐波成分并组成特征向量。然后用DBN分类器对扰动信号进行分类识别,DBN方法比常用的分类方法具有更高的分类准确率和更短的训练时间。通过应用于现场实测扰动数据表明:所提出的方法适用于多种类型的电能质量扰动检测,在少样本情况下具有优越的分类性能。In order to meet the requirements of accurately classifying power quality disturbances,a method for power quality disturbance classification is proposed based on maximal overlap discrete wavelet transform(MODWT)and deep belief networks(DBN).Initially,a reliable power quality disturbance detection algorithm is proposed by using MODWT.This algorithm can obtain the power quality transient events beginning and ending time accurately without setting detection threshold,from whose results the voltage harmonic components of power quality transient events are extracted and used to form feature vector.Then,DBN,as a classifier,is used to classify power quality disturbances.Comparing with common classification method,DBN has higher accuracy and shorter training time.The test results based on power grid field data show that the proposed method is suitable for detecting various types of power quality disturbances,and it is characterized by high recognition correctness with insufficient training samples.

关 键 词：电能质量 极大重叠离散小波变换 深度置信网络 分类识别 

分 类 号：TM72[电气工程—电力系统及自动化]