An analysis of power grid alarm data with clustering algorithms and the associated processing model

Abstract

Alarm data of power grids can be classified into various groups according to different criteria, therefore making it extremely hard to be understood. A clustering algorithm based on timestamps and processed keywords is proposed to solve these problems. First, the K-Means algorithm is conducted to cluster the latest occurrence time. The original data set is divided into multiple groups. Secondly, the secondary K-Means clustering is also performed based on the start time of the alarm data for further corrections. Next, we extract the keywords features, and then vectorize them. Then we apply the density-based spatial clustering of applications with noise (DBSCAN) algorithm for distinguishing these keyword features. Finally, results with both the K-Means and the DBSCAN methods are integrated and presented, and their descriptions on correlation are provided. Moreover, classifications of the alarm events in the original data are also specified. The experimental results show that the proposed model can achieve an average compression rate of 23.62% and an average accuracy rate of 93.41%. Consequently, the proposed model can effectively improve the ability to visualize the data and reduce the complexity of alarm data.