Classification in complex systems through negative recognition

In this work, the applications of an approach that is based on establishment of class membership to diagnosis of complex system faults are reported. The adaptive recognition to achieve the classification is based on discovery of pattern features that make them distinct from objects belonging to different classes. In contrast to most systems for fault identification and diagnosis, which depend on heuristic rules, this approach does not resort to any heuristic rule. Consequently, it is more appropriate for diagnosis of faults in large and complex systems. To facilitate the evaluation of the ensuing scheme, results of diagnosis for a large power system, based on data provided by its protection simulator, are also reported. Those results clearly demonstrate that, after proper training, with minimal supervision, fast and successful diagnosis of all faults can be achieved.