Uma arquitetura híbrida para sistemas cognitivos e sua aplicação em prognóstico de falhas em transformadores de potência

Abstract

This project proposes a framework architecture to export services and facilities for the implementation of knowledge-based systems. The framework utilizes concepts from embryonic, connectionist and evolutionary approaches. Additionally, we map correlations between computational systems with natural cognition, knowledge formation and representation. This way, this framework facilitates the implementation of generic intelligent systems. The connectionist approach employs ensemble learning for its potential in achieving higher accuracy and robustness. Ensemble learning can combine together different classifiers based on machine learning that are available in the literature and statistical methods. The symbolic approach employs an evolutionary method to permit the knowledge dynamics. The main contribution of the project relies in the utilization of symbolic and connexionist approaches to cover all the knowledge representation aspects. As result, was implemented a system able to generate fault prognosis in power transformers.