Independent component analysis for feature reduction in critical clearing time estimation

Critical clearing time (CCT) is an important parameter used in transient stability assessment of power systems. In this paper, only pre-fault generator voltage magnitudes and active powers are used as inputs to a multilayer feedforward neural network (MFNN) trained using Gauss-Newton approximation for Bayesian regularization to estimate the CCT in an optimal power flow framework. Principal component analysis (PCA) and independent component analysis (ICA) are utilized to reduce the dimension of these features. Simulations performed on the New England 10-machine 39-bus system show that ICA outperforms PCA and gives a smooth degradation of MFNN performance as the number of input features are reduced.