Indirect PCA Dimensionality Reduction Based Machine Learning Algorithms for Power System Transient Stability Assessment

Large dimensionality of data makes the training samples sparse within each dimension, and it will weaken the effectiveness of machine learning methods while the number of training data is not sufficient. Dimensionality reduction is a natural way to improve the performance of machine learning applications. In this paper, an indirect principal component analysis (PCA) algorithm to reduce data dimensions is proposed. The data dimensions on which the projection values on PCA base have the larger difference to distinguish samples with different marks, are selected to retain, and the other dimensions are discarded. The dimensionality reduction algorithm is designed to coordinate with machine learning methods, and its advantages and effectiveness over traditional methods are demonstrated through the power system transient stability assessment tests.