Data mining & pattern recognition of voltage sag based on K-means clustering algorithm

With the increasing demands of power supply, the electric power quality especially the voltage sag deserves more concerns. This paper presents an approach of K-means clustering analysis algorithm to classify and recognize the voltage sag from the measured historical data of large-scale grid in Shenzhen, China. The distances among different sag incidents in distribution diagram are calculated first. When some distances are nearer, a cluster center which is called centroid can be set to represent these incidents. Then the centroid amounts and locations are determined based on iterative updating method. The sag amplitude and duration time reflected by these centroids can be regarded as the voltage sag characteristics of similar substations, which will represent the operation condition and find out the weak link of whole power systems. Thus the algorithm converts the complicated and disordered sag incidents into some typical sag models, which provides the theoretical evidence for simplifying analysis and practical management of voltage sags.