基于K-means++算法的日负荷曲线聚类分析

Cluster Analysis of Daily Load Curves Based on K-means++ Algorithm

摘要:

k-means聚类算法因计算过程简单、收敛速度快，被广泛应用于负荷特性分析。然而，k-means聚类算法的聚类数难以选择，且随机的初始化质心选择易导致收敛速度慢和陷入局部最优的问题。因此，提出一种基于k-means++算法的日负荷曲线聚类分析方法，利用启发式随机播种方式选取初始质心，基于肘部法则，利用聚类评价指标度量聚类的密集度和分离度，综合评定确定最佳聚类数，避免初始质心的随机性影响聚类质量。算例表明，所提方法在聚类质量方面表现出较高水平，能够提供较为合理的负荷分类，有助于掌握用户的负荷特性。

Abstract:

The K-means cluster algorithm is widely used for the analysis of load characteristics due to its simple calculation process and fast convergence rate. However, selecting the number of clusters in K-means algorithm is challenging, and the random initialization of cluster centers can lead to slow convergence rate and local optima issues.A cluster analysis method for daily load curves based on K-means++ algorithm is proposed. It selects the initial cluster centers using heuristic random seeding method, combines the elbow method and clustering evaluation metrics to measure the density and separation of clusters and comprehensively determines the optimal number of clusters, which mitigates the impact of randomness of initial cluster centers on clustering quality. The case studies show that the proposed algorithm can achieve high clustering quality, provide a reasonable user classification and facilitate a comprehensive understanding of load characteristics.