Regras extraídas de um espaço de entrada de alta dimensão através da adição de tributos para a obtenção de um novo método de análise de faltas incipientes em transformadores

Abstract

For diagnosis of incipient faults in power transformers and other power system elements are usually utilized the concentrations of dissolved gases present in the insulation oil, through DGA-based (Dissolved Gas Analysis) methods, like the Duval Triangle and others, which are described in IEEE C57.104 and in IEC 60599 standards. However, they present some limitations such as the existence of non-decision areas and erroneous results. In this paper, will be described a new methodology to improve the analysis of incipient faults based on rules extracted from a high-dimension space (21 attributes), formed by the gases concentrations and some of their interrelations. Utilizing a decision tree method (C4.5), a set of interpretable rules is extracted from high-dimension input space. Databases known in the DGA technical literature are adopted to analyze the proposed approach, with the new method being compared with standard methods. Considering data test folders in the performed 10-folder cross-validation statistical analysis, the extracted rules show greater accuracy with an error in the diagnosis of incipient faults of 6.25% against 18.75% from the Triangle method in the worst case. Were also performed reliability tests, comparing the new method with methods cited in the standards.