New Method for Power System Dynamic Stability Analysis Based on a Novel Unsupervised Clustering Algorithm

A new method for power system dynamic stability assessment (DSA) is proposed in this paper. The approach is applied the unsupervised concept to clustering the training set and only used the power system steady state quantities. In this technique, the clustering strategy is adaptive to different data shapes, which utilizes residual value index to objectively estimate whether a subspace contains data structure information and the hierarchical process can guarantee our algorithm effective for data set with different figures. The cluster representation is only dependent on external samples, thus can be easily stored and used to build a classification algorithm. Besides, our method accepts new training samples conveniently by only analyzing those new sample points on the base of the obtained clustering results. Application results on power system DSA problem show its merits as an unsupervised clustering algorithm and thus can be treated as a tool for DSA.