采用深度学习和多维模糊C均值聚类的负荷分类方法  

Load Classification Method Using Deep Learning and Multi-dimensional Fuzzy C-means Clustering

作　　者：石亮缘 周任军[1] 张武军 余虎 李彬[1] 王珑 SHI Liangyuan;ZHOU Renjun;ZHANG Wujun;YU Hu;LI Bin;WANG Long(Hunan Province Collaborative Innovation Center of Clean Energy and Smart Grid,Changsha University of Science and Technology,Changsha 410114,China;Hunan Electric Power Design Institute Co.,Ltd,China Energy Engineering Group,Changsha 410007,China)

机构地区：[1]湖南省清洁能源与智能电网协同创新中心(长沙理工大学),长沙410114 [2]中国能源建设集团湖南省电力设计院有限公司,长沙410007

出　　处：《电力系统及其自动化学报》2019年第7期43-50,共8页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金资助项目(71331001,51277016)

摘　　要：为了对日趋海量的负荷数据进行有效地分类处理,提出一种采用深度学习和多维模糊C均值聚类的负荷分类方法。采用深度学习中的卷积自编码器CAEs堆叠形成深度卷积自编码网络,通过训练实现对输入的典型日负荷曲线集进行特征分层提取和降维处理。计及低维特征序列的数值维度和趋势维度,将数值序列的欧氏距离与趋势序列的改进动态时间弯曲距离相结合为多维相似性距离,作为新的相似性指标,提出一种多维模糊C均值聚类算法,用以对特征序列进行聚类分析。算例分析结果表明,所提出的方法在数据特征提取降维、负荷分类有效性、稳定性及聚类效率等方面具有较大优势,可为需求侧管理项目选择、电价制定、负荷管理优化等提供有效参考。To effectively classify the increasingly massive load data,a load classification method using deep learning and multi-dimensional fuzzy C-means clustering is proposed. A deep convolutional auto-encoding network is formed by convolution auto-encoders(CAEs)stacking in deep learning. Through training,layered feature extraction and dimensionality reduction are performed on the typical daily load curve set of input. By taking the numerical and trend dimensions of the low-dimensional feature sequence into account,the Euclidean distance of numerical sequences is combined with the improved dynamic time warping distance of trend sequences,thus forming a multi-dimensional similarity distance as a new similarity indicator. Moreover,a multi-dimensional fuzzy C-means(FCM)clustering method is proposed to perform clustering analysis of feature sequences. The results of a numerical example show that the proposed method has advantages in data feature extraction and dimensionality reduction,effectiveness of load classification,stability,and clustering efficiency,which can provide effective reference for the selection of demand-side management projects,electricity pricing,load management optimization,etc.

关 键 词：深度学习 卷积自编码器 多维特征 模糊C均值聚类 负荷分类 

分 类 号：TM734[电气工程—电力系统及自动化]