Research of the FGA-ANN method for transformer fault diagnosis based on the dissolved gas analysis

Dissolved gas analysis has been used as a diagnostic method to determine the conditions of transformers for a long time. The criteria used in dissolved gas analysis are based on crisp value norms. Due to the dichotomous nature of crisp criteria, transformers with similar gas-in-oil conditions may lead to very different conclusions of diagnosis especially when the gas concentrations are around the crisp norms. To deal with this problem, gas-in-oil data of failed transformers were collected and treated in order to obtain the membership functions of fault patterns using a grey relational analysis method. All crisp norms were transformed into mapping rules. In this paper, the novel method of fuzzy genetic algorithm-artificial neural networks (FGA-ANN) was applied to transformer fault diagnosis instead of the ratio method. The novel method combined GA and ANN, during genetic algorithm´s optimized, crossover rate and mutation rate were adjusted dynamically by fuzzy control. The treated data of the model samples were operated by FGA-ANN and a group of weighs and biases were found. Finally examples were given. Compared to the other traditional method, the results have demonstrated the robustness of the method.