Comparison of feature selection techniques for ANN-based voltage estimation

Fast estimation of bus voltage magnitude is essential for security monitoring and analysis of power system. An approach based on a parallel self-organising hierarchical neural network (PSHNN) is proposed to estimate bus voltage magnitudes at all the PQ buses of a power system in an efficient manner. PSHNN is a multi-stage neural network in which stages operate in parallel rather than in series during testing. The revised back-propagation algorithm is used for learning input non-linearities along with forward-backward training of stage neural networks. A method based on Euclidean distance clustering is proposed for feature selection. Effectiveness of the proposed method is compared with two existing methods of feature-selection entropy based and angular distance based clustering methods for bus voltage magnitude estimation at different loading conditions in the IEEE 30-bus system and a practical 75-bus Indian system. The PSHNN based on Euclidean distance based clustering method is found to be superior in terms of training time and error performance. (C) 2000 Elsevier Science S.A. All rights reserved.