Hybrid supervised and unsupervised neural network approach to voltage stability analysis

This paper presents a hybrid neural network approach to voltage stability analysis. Essentially, the proposed hybrid neural network consists of a Kohonen network and a multi-layer feed-forward network. By using voltage collapse margin method (VCM) and singular value decomposition method (SVD), the hybrid neural network is trained to identify voltage weak buses/areas and to evaluate the loadability of power systems in terms of voltage stability. Moreover, the Kohonen neural network works as a front end to cluster input patterns with similar features of operating conditions and hence the generalization capability of the multi-layer feed-forward network has been improved significantly. The effectiveness of the proposed network has been demonstrated on the IEEE 57-bus test system and the results are very encouraging. (C) 1998 Elsevier Science S.A. All rights reserved.