Fuzzy and Genetic Approach to Diagnosis of Power Transformers

Abstract The paper describes algorithms enabling construction of an expert system to diagnose technical condition of an oil transformer on the basis of DGA (Dissolved Gas Analysis), i.e. the results of chromatography of gases dissolved in transformer oil. Two soft-computing approaches are presented. Firstly, the diagnosis is made by a fuzzy controller making use of the rules resulting from fuzziness introduced in the IEC code, which is frequently applied to transformer diagnosing. To train the controller, the system exploits genetic algorithms. Secondly, the method determining regions within the space of DGA-results which are relevant to specific diagnostic statements. Fuzziness helps to define the regions while genetics - to reduce the number of rules. In both cases, numerical results compare well with the actual diagnosis performed by human-expert.