Metodologia não Supervisionada de Diagnostico da Integridade em Transformadores de Distribuição Através da Detecção de Anomalias

This paper presents an unsupervised method of diagnosis of the health of distribution transformers. The kernel of this methodology is a generic adaptive model that uses the parameters measured in the transformer as a signature. A strategy is introduced to detect abrupt changes. Furthermore, operating zones for the transformer are proposed. This methodology runs autonomously and continuously updates its decision model by implementing an unsupervised recursive learning algorithm. The results of the evaluation of the proposed methodology are carried out by monitoring two distribution transformers in service.