Transformer User Identification Method Based on Abnormal Point Detection and Large Data Clustering

2019 5th International Conference on Energy Materials and Environment Engineering

能源学;生态环境科学

Liu, Chen^1 ; Wang, Junping^1 ; He, Zhongxiao^2 ; Wu, Meng^1 ; Bai, Jialing^1 ; Lian, Libo^2

State Grid Sichuan Electric Power Corporation Metering Center, Chengdu

610000, China^1

Tsinghua Sichuan Energy Internet Research Institute, Chengdu

610000, China^2

关键词: Application scenario;    Correlation coefficient;    Determining the number of clusters;    Identification accuracy;    Improved K-Means algorithm;    Intelligent management;    Sample similarities;    User identification;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/295/5/052036/pdf
DOI  :  10.1088/1755-1315/295/5/052036

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

Solving the mismatch of transformer and users is a key step to promote the intelligent management of distribution network. The rapid popularization of big data technology makes it possible to achieve low-cost and high-efficiency transformer user identification. Since load transferring may occur during the operation of distribution network, the user data to be identified may be mixed with the operation data of other transformer users. If the users' data of other transformers is not eliminated, the identification accuracy will be greatly reduced. In this paper, a transformer user identification method based on loacal outlier detection and improved k-means algorithm is proposed. Firstly, the analysis data is preprocessed by the local factor algorithm, the user data which does not belong to the transformer to be analyzed is eliminated. According to the characteristics of application scenarios, improved k-means algorithm is proposed, including determining the number of clusters, initial centroid and selecting correlation coefficient as the index of evaluating sample similarity. Finally, the k-means algorithm is used to cluster the pre-processed data to accurately identify the transformer users. Example results show that the proposed method can effectively improve the accuracy of transformer user identification and maintain high stability in different data environments.