基于降噪自编码的电能质量扰动识别  

Classification of power quality disturbances based on denoising autoencoder

作　　者：肖露欣[1] 李增祥 马建 陈克绪 吴建华 XIAO Luxin;LI Zengxiang;MA Jian;CHEN Kexu;WU Jianhua(School of Information Engineering,Nanchang University,Nanchang 330031,China;Gongqing College,Nanchang University,Jiujiang332020,China;State Grid Jiangxi Electric Power Company Electric Power Research Institute,Nanchang 330096,China)

机构地区：[1]南昌大学信息工程学院,江西南昌330031 [2]南昌大学共青学院,江西共青城332020 [3]国网江西省电力公司电力科学研究院,江西南昌330096

出　　处：《南昌大学学报（理科版）》2017年第6期591-595,共5页Journal of Nanchang University(Natural Science)

基　　金：国家自然科学基金资助项目(61662047);江西省研究生创新专项资金资助项目(YC2015-S037)

摘　　要：针对目前电能质量扰动识别时特征提取不充分,造成识别精度不高的问题,引入了降噪自编码算法。降噪自编码算法起源于自动编码算法,两者都属于深度学习算法。其中,自动编码算法已经被应用于电能质量扰动识别,并取得了一定的成果。但是,自动编码算法对含噪声干扰的电能质量扰动信号的识别精度还不是很理想。本文采用降噪自编码算法,将克服这一问题。首先对无噪声的扰动信号用噪声进行"破坏",然后用带噪声信号去重构原始信号,得到扰动信号波形的固有特征,最后通过BP神经网络分类器对整个网络进行微调,得到最后用于分类的特征样本。该方法降低了传统特征提取算法对特征选取不当,造成分类识别精度不理想的风险,并在一定程度上提高了含噪声的电能质量扰动信号的识别精度。仿真结果表明,该方法在识别含噪声的电能质量扰动信号上有很大的优势。In view of the currently insufficient feature extraction and unsatisfied classification of power quality disturbances(PQDs),this paper employs a denoising autoencoder(DAE)based algorithm to improve this situation.The DAE originates from the autoencoder(AE),and both of them belong to the framework of deep leaning.The AE algorithm has been used in classification of PQDs successfully and made some progress.However,the classification accuracy of noisy PQDs is not ideal by simply using AE algorithm.In this paper,the use of DAE algorithm will address this issue.First,the original PQD signals are corrupted by noise,and then,the intrinsic features of PQDs waveforms are obtained by reconstructing the original PQDs from the corrupted signals,finally,by using a BP neural network to fine tune the whole network model,feature samples suitable for classification are available.This proposed algorithm reduces the risk of unsatisfied classification accuracy due to improper feature extraction of traditional methods for noisy PQDs.Simulation results show that the proposed algorithm has great advantage in classification of noisy PQDs over state-of-the-art methods.

关 键 词：电能质量 特征提取 降噪自编码 噪声 分类识别 

分 类 号：TM711[电气工程—电力系统及自动化]