基于欧氏动态时间弯曲距离与熵权法的负荷曲线聚类方法

Load Curve Clustering Method Based on Euclidean Dynamic Time Warping Distance and Entropy Weight

为了改善目前负荷建模中聚类方法相似度衡量不准确及聚类结果质量较差的问题,综合运用k-means及熵权法原理,提出一种基于欧氏距离与动态时间弯曲距离的日负荷曲线聚类方法。首先,采用欧氏距离与动态时间弯曲距离分别衡量日负荷曲线的整体分布特性、局部动态特性与整体动态特性。然后,引入熵权法自适应配置3种特性的权重系数。最后,采用k-means聚类算法,以所提相似度衡量方法为依据,对用电日负荷曲线进行聚类。算例对某省区电网典型用户的日负荷曲线展开聚类分析,结果表明所提方法相似度衡量指标合理,且在聚类质量、鲁棒性等方面具有一定的优越性,可以真实反映该地区的用户用电特性,满足在线负荷建模的应用需求。

In order to improve the accuracy of similarity measurement and the quality of clustering results in current load modeling,a daily load curve clustering method based on Euclidean distance and dynamic time warping distance is proposed by using the principle of k-means and entropy weight method. Firstly, Euclidean distance and dynamic time warping distance are adopted to measure the overall distribution characteristics, local dynamic characteristics and overall dynamic characteristics of the daily load curves. Secondly, entropy weight method is introduced to adaptively configure the weight coefficients of these three characteristics.Finally, k-means clustering algorithm is used to cluster the daily load curves based on the proposed similarity measurement method. The clustering analysis of daily load curves of typical consumers in a provincial power grid is made. The results show that the similarity measurement indices selected in the proposed method are reasonable, and the method has certain advantages in clustering quality and robustness, which can truly reflect the power consumption characteristics of consumers in this area and meet the application requirements of online load modeling.