An Adaptive Two-Stage Clustering Algorithm for Load Series to Maximize Inter-Cluster Distances

The load characteristics of modern power systems are becoming more complex due to the increased penetration of distributed PV generation and EV charging, making it difficult to distinguish probability distributions for clustering purposes. In this paper, an adaptable two-stage clustering algorithm is proposed. Firstly, the regions with high volatility are automatically extracted as regions of interest (ROI) for the statistical features of different load data. In the first stage, the improved Davies-Bouldin index (DB index) is used to maximize the distance between clusters to find general patterns, and the number of clustering iterations and the optimal number of clusters is automatically selected according to this criterion. In the second stage, the clusters with large intra-cluster distances are then re-clustered to be refined and to extract special patterns. The method was tested on real EV charging loads and actual regional load data in East China. Finally, the interpretability of EV load clustering results is analyzed based on statistical features, which demonstrates the effectiveness of the method.