| 引用本文: | 屈相帅,段斌,尹桥宣,晏寅鑫,钟颖.基于稀疏自动编码器深度神经网络的电能质量扰动分类方法[J].电力自动化设备,2019,39(5): |
| QU Xiangshuai,DUAN Bin,YIN Qiaoxuan,YAN Yinxin,ZHONG Ying.Classification method of power quality disturbances based on deep neural network of sparse auto-encoder[J].Electric Power Automation Equipment,2019,39(5): |
|
| 摘要: |
| 针对智能电网日益突出的电能质量扰动问题，提出了一种基于稀疏自动编码器(SAE)深度神经网络的电能质量扰动分类方法。利用SAE对电能质量扰动原始数据进行无监督特征学习，自动提取数据特征的稀疏特征表达；通过堆栈式稀疏自动编码器(SSAE)进行逐层学习，获得电能质量扰动数据的深层次特征；将其连接到softmax分类器进行微调训练，并输出电能质量扰动事件分类结果。利用已添加高斯白噪声的数据对SSAE进行训练，以提高其特征表达的抗噪声能力。仿真结果表明，所提方法能够准确地识别包含2种复合扰动在内的9种电能质量扰动信号，并且具有很好的鲁棒性。 |
| 关键词: 电能质量 扰动分类 特征提取 扰动识别 稀疏自动编码器 深度学习 |
| DOI：10.16081/j.issn.1006-6047.2019.05.023 |
| 分类号:TM761 |
| 基金项目:国家自然科学基金资助项目(61379063) |
|
| Classification method of power quality disturbances based on deep neural network of sparse auto-encoder |
|
QU Xiangshuai1,2, DUAN Bin1,2, YIN Qiaoxuan1,2, YAN Yinxin2, ZHONG Ying2
|
|
1.Key Laboratory of Intelligent Computing & Information Processing of Ministry of Education, Xiangtan University, Xiangtan 411105, China;2.Cooperative Innovation Center for Wind Power Equipment and Energy Conversion, Xiangtan University, Xiangtan 411105, China
|
| Abstract: |
| Aiming at the increasingly prominent power quality disturbances in smart grid, a classification method of power quality disturbances based on deep neural network of SAE(Sparse Auto-Encoder) is proposed. The unsupervised feature learning is carried out for the original data of power quality disturbances by using SAE and the sparse feature expressions of data features are extracted automatically. The deep features of power quality disturbance data are acquired by learning layer by layer based on SSAE(Stack Sparse Auto-Encoder). The deep features are connec-ted to the softmax classifier for fine-tuning training to obtain the classification result of power quality disturbances. SSAE is trained based on the data added with Gaussian white noise to improve the anti-noise ability of its feature expression. Simulative results show that the proposed method can accurately identify nine kinds of power quality disturbance signals including two types of complex disturbances, and have good robustness. |
| Key words: power quality classification of disturbances feature extraction disturbance identification sparse auto-encoder deep learning |
