Combination of rough set theory and artificial neural networks for transient stability assessment

Power system transient-stability assessment (TSA) based on pattern recognition techniques can usually be treated as a two-pattern classification problem separating the stable class from the unstable class. Two underlying problems are (1) selecting a group of effective features (attributes), and (2) building a pattern classifier with high classification accuracy. This gaper proposes to combine the rough set theory (RST) with a back-propagation neural network (BPNN) for TSA, including feature extraction and classifier construction. First, through discretization of the initial input attributes, the inductive learning algorithm based on RST is employed to reduce the input attribute set. Then, a BPNN using a semi-supervised learning algorithm is used as a 'rough classifier' to classify the system stability into three classes-stable class, unstable class and indeterminate class (boundary region). The introduction of the indeterminate class provides a feasible way to reduce misclassifications, and the reliability of the classification results can hence be greatly improved. The validity of the proposed approach for both feature extraction and removing misclassifications of BPNN-based TSA is verified by the 10-unit New England power system.