Genetic-based Feature Selection Applied to Neural Networks for Security Boundary Visualization

Knowledge of security boundaries is becoming more and more important as transmission lines are operating closer and closer to their capacities in today's deregulated environment. On-line visualization of security boundaries using intelligent techniques provides an eeective methodology for this problem. Feature selection , i.e., critical parameter selection, is one of the most essential techniques of this methodology. This paper presents an automatic feature selection approach called GANN (Genetic Algorithm based Neural Network feature selection) based on genetic algorithm (GA) to select critical parameters applied to neural networks for boundary visualization. The resulting boundaries can be presented to operators for on-line use. A demonstration shows that the boundary accuracy, obtained from neural network trained using the selected features, is good.