A fast screening method for transient stability considering multi-swing step-out using pattern recognition with machine learning and clustering

Recently, online stability monitoring systems have become more important in response to the increasing complexity of power systems. Moreover, there has been a concern about multi-swing step-out due to the Japanese longitudinal power system. In this paper, a fast screening method is proposed considering multi-swing step-out using PCA (principal component analysis). In the proposed method, computers learn patterns of PCA in transient stability data as a form of library. In order to reduce the number of data in the library, k-means method, one of the partitioning-optimization clustering methods, is applied to extract features in the data. In addition, Gaussian mixture model is also applied to extract the feature from a different perspective. Simulations for the proposed method are performed using the IEEJ 10 machine 47 bus system to confirm the validity of the screening method.