基于受限玻尔兹曼机的电能质量复合扰动识别  

Classification on mixed disturbances of power quality based on restricted boltzmann machine

作　　者：马建[1] 陈克绪 肖露欣[2] 吴建华[2] 

机构地区：[1]国网江西省电力科学研究院,江西南昌330096 [2]南昌大学信息工程学院,江西南昌330031

出　　处：《南昌大学学报（理科版）》2016年第1期30-34,共5页Journal of Nanchang University(Natural Science)

基　　金：国家自然科学基金资助项目(61162014);江西省自然科学基金资助项目(20122BAB201029)

摘　　要：针对目前电能质量混合扰动识别精度不高的问题,引入了受限玻尔兹曼机(RBM)算法。RBM是深度学习的一种新颖算法,在语音识别、机器视觉和图像恢复等领域已取得了很好的应用成果,但在电能质量复合扰动识别上尚未涉及。区别于传统算法提取特征的方式,深度网络通过提取波形的固有抽象特征,克服了人工特征选择的缺陷以及传统神经网络训练时收敛速度慢、容易限于局部最优的缺点。复合扰动信号经过深度网络自动获得特征参数,再经过分类器进行分类识别。实验验证该算法在电能质量复合扰动识别上可以达到很高的性能,优于传统的识别方法。In view of the currentlyunsatisfied classification of mixed power quality disturbances,we employedthe Restricted Boltzmann machine algorithm（RBM）in this paper.The RBM is a novel algorithm in deep learning.It has achieved some good results in speech recognition,machine vision and image restorationetc.But it has not been involvedin the classification onmixed disturbances of power quality yet.Different from the explicit features extraction bytraditional algorithms and by extracting intrinsic abstract characteristics of waveform,the deep network mightovercome the defect of the artificialfeatures selection andavoid the shortcoming of slow convergence and vulnerability to fall into a local minimum intraining of the traditional neural network.When the mixed disturbance signals wereapplied into the deep network,the feature parameterswouldbe automatically obtained and then classified by the classifier.Our simulation results showed that the RBM algorithm could achieve a better performance than traditional classification algorithms in classifyingthe mixed power quality disturbances.

关 键 词：电能质量 复合扰动 受限玻尔兹曼机 深度学习 分类识别 

分 类 号：TM711[电气工程—电力系统及自动化]