基于多维尺度分析和改进K-means的台户关系辨识方法

Identification Method for Station-user Relationship Based on Muti-Dimensional Scaling and Improved K-means

智能电网的快速发展对配电网侧的精细管理提出了更高的要求。然而,终端用户难以和台区管控的配电变压器准确配准,使得台区智能化管控的多个高级应用难以推进。配电网运行过程中的海量数据,尤其是电压数据的变化趋势能够反映配网的线变关系。为此,介绍了基于多维尺度分析(muti-dimensional scaling,MDS)和改进K-means的台户关系辨识方法。首先,通过MDS算法对台区变压器低压侧所采集到的电压数据进行降维处理,从而降低整体算法计算量,提高算法效率;并根据特定应用场景对K-means算法做了如下改进:包括以变压器总相数确定聚类个数、以变压器出口电压作为初始聚类中心,并用相关系数作为衡量相似度的标准,从而提升算法的整体准确性。算例分析结果表明:所提方法能够有效提升台区用户辨识的准确度,在低密度数据或问题复杂度增加的情况下依旧能够保持较高的识别精度,且效果稳定。

Rapid development of smart grids puts forward higher requirements for fine management at the distribution network side.However,accurate rectification between the end user and the distribution transformer controlled by the station area is hard to achieve,so that it becomes difficult to use a number of advanced applications of the intelligent management and control of the station area.The mass data in distribution network operation,especially those about voltage variation tendency,can reflect variations between lines of the distribution network.For this reason,a method for identifying the station-area relationship based on muti-dimensional scaling(MDS)and improved K-means was introduced in this paper.Firstly,the voltage data collected at the low-voltage side of the transformer of the station area was processed in the MDS algorithm to reduce the overall algorithm computation and improve the efficiency of the algorithm.Furthermore,depending on specific application scenarios,the K-means algorithm was improved in following ways:the number of clusters was determined through the total number of phases of the transformer,the transformer outlet voltage was taken as initial clustering center,and the correlation coefficient was used as criterion for similarity measurement so as to improve the overall accuracy of the algorithm.The results of case studies indicated that the proposed method could effectively improve the accuracy of user identification in the station area,and could still maintain a quite high recognition accuracy in the case of low-density data or higher complexity.