Improvement of classification accuracy by using enhanced query-based learning neural networks

Abstract

An enhanced query-based learning neural network is proposed for the dynamic security control of power systems. Compared to conventional neural network, the enhanced query-based learning provides a classifier at lower computational cost. This methodology requires asking a partially trained classifier to respond to the questions. The response of the query is then taken to the oracle. An oracle is responsible for providing better quality of training data. The regions of classification ambiguity will thus be narrowed. It can be seen that the proposed method is intrinsically different from learning by randomly generated data. With only a small amount of additional complexity, the enhanced query-based neural network approach greatly increases the classification accuracy of neural networks.