Artificial Neural Network Applied to Differential Protection of Power Transformers

This paper presents a differential protection algorithm for power transformers based on artificial neural network (ANN). The electrical systems studied were modeled and simulated using the alternative transients program software. The proposed protection scheme was applied to three transformers with rated powers of 100, 18 and 40 MVA. The following electromagnetic phenomena were simulated inrush, sympathetic inrush, overexcitation, saturation of current transformers and internal and external faults in the transformer protection zone. For each transformer, a database was generated with 1210 files for training and testing the ANN. This computational intelligence tool was implemented in MATLAB® software. The obtained results show that the developed methodology presented superior performance when compared to the conventional differential protection, proving to be a promising solution in this type of application. The analyzed real cases had 100% success rate.