Discriminant function for insulation fault diagnosis of power transformers using genetic programming and co-evolution

This paper presents a discriminant function model for insulation fault diagnosis of power transformers using genetic programming and co-evolution. Our model uses two evolutionary algorithms: a genetic programming algorithm evolving a population of discriminant functions and an evolution strategy algorithm evolving a population of division points. Global search abilities of evolutionary algorithm enable the two populations co-evolve, so that the final result of the co-evolutionary process is a discriminant function and a set of division points which are well adapted to each other. The relationships among the concentrations of dissolved gases in transformer oil with corresponding fault types are featured by discriminant function and division points. The proposed method has been tested on the real diagnostic records and compared with conventional IEC method, fuzzy diagnosis system and artificial neural networks. The results show that our method has the advantage of existing methods in diagnosis accuracy.