基于大数据技术的地区电网负荷特征分析  

Analysis of Load Characteristics of Regional Power Grid Based on Big Data Technology

作　　者：刘鹏远 LIU Pengyuan(Xi’an Power Supply Company,Xi’an 710000)

机构地区：[1]国网陕西省电力公司西安供电公司,陕西西安710000

出　　处：《电力安全技术》2020年第12期18-23,共6页Electric Safety Technology

摘　　要：针对传统机器学习算法存在的问题,研究了在地理分布式情景下的用户负荷特征聚类算法。提出了在单节点集中数据集下生成局部聚类模型时采用的PCA-负荷指标特征融合算法以及集中式K-means算法。针对多节点地理分布式数据集,构建了考虑特征迁移的分布式聚类模型参数共识方法框架,提出了将所有节点的局部聚类模型通过参数共识得到全局聚类模型的分布式K-means算法,同时考虑了模型参数与数据集特征的可迁移性。通过某地区用户负荷数据对模型进行了验证。结果显示,分布式K-means能利用全局信息、考虑不同区域的差异来更好地识别典型用电曲线,并且算法具有较好的可迁移性。Aiming at the problems existing in traditional machine learning algorithms, this paper studies a user load characteristic clustering algorithm in geographically distributed scenarios. Firstly, the paper proposes a PCA-load index feature fusion algorithm and a centralized K-means algorithm, which are used to generate a local clustering model under a single node centralized data set. Secondly, for the multi-node geographically distributed data set, it constructs a parameter consensus method framework for distributed clustering model considering feature transferring, proposes a distributed K-means algorithm that obtains the overall clustering model by using the local clustering model of all nodes through parameter consensus, meanwhile, it considers the transferability of model parameters and data set features. Finally, the paper verifies model with customer load data from a certain region. The results show that distributed K-means can use overall information with considering differences in different regions to better identify typical power consumption curves, and the algorithm has better transferability.

关 键 词：地理分布 负荷特征分析 分布式聚类 迁移学习 大数据技术 

分 类 号：TM74[电气工程—电力系统及自动化]