The dynamic system stability analysis with combined unsupervised and supervised learning methods

Abstract

This paper deals with the stability analysis of dynamic systems with unsupervised and supervised learning of an artificial neural networks. Several different artificial neural networks configurations were tested for determining the stability level of the system whose topology varies. Stability analysis is achieved by tracking changes in the dominant eigenvalues of the system which are the output of the trained multi- layered neural network. Testing was done of the possibility of using a multi-layered neural networks which had undergone supervised learning to define the stability level of the system on the basis of the complete or reduced state vector. The reduced state vector is obtained by extraction of the characteristic state coordinates by means of unsupervised learning. Results from the test power system indicate a wide variety of possibilities for applying this approach in complex dynamic systems.