Improving the diagnosis of incipient faults in power transformers using dissolved gas analysis and multi layer perceptron

In this work, a ten year gas production records, obtained from Rodgers ratio codes for gas concentration of H
₂
, CH
₄
, C
₂
H
₆
, C
₂
H
₄
and C
₂
H
₂
, in a 795 MVA 20/400 kV rated power transformer are used in MLP for improving the diagnosis accuracy of incipient faults (partial discharge, thermal faults, low and high energy discharges, etc...). Numerical modelling of DGA-sourced data was achieved using MATLAB platform. Performance validation for the neural network was assessed in terms of the regression fit (R), the MSE and the MAPE. Results show 2.51 % as optimal value of MAPE which indicate a reasonable accuracy of ANN technique in the prediction of incipient faults.