Skip to Main Content
In this paper, the authors present a pattern-learning/recognition approach for dynamic security classification using neural networks with a limited number of input data. The input is a set of data representing the precontingency power system state (voltages, angles, etc.), and the output is the possible system status (stable/unstable) after contingency. Data clustering is applied to reduce the number of input representing the cases. The reduced input data are then used to train the neural network that learns the input patterns for a possible post-contingency status. The overall accuracy of the classification is considered to be reasonable for a practical-scale power system application.