ACTA

Discharge Currents Discrimination Technique Based on Multi-Linear Regression Line and Artificial Neural Networks for Power Transformers Diagnosis

F. Benyahia^a, A. Nacer^a, H. Moulai^a, F. Aberkane^a and A. Beroual^b
^aLaboratory of Electrical and Industrial Systems, FEI, USTHB, BP 32 Bab Ezzouar, Algiers, 16311, Algeria
^bEcole Centrale de Lyon, Ampere CNRS UMR 5005, 36 av. Guy de Collongue, 69134 Ecully, France

Full Text PDF

The proposed work will be consecrated to the study of positive pre-breakdown currents triggered in mineral transformer oil under 50 Hz alternating overvoltage. Since negative currents are recorded in low rates and for higher voltage levels than positive ones, only the latter will be prior taken into consideration. Both streamer propagation and arc discharge current types are identified and are used in the training process of an artificial neural network and the multi-linear regression line of these currents in order to develop a complementary diagnosis tool that can serve as an on-line transformer protection. More successful results than those obtained by other developed techniques are expected.

DOI: 10.12693/APhysPolA.123.250
PACS numbers: 77.22.Jp, 52.80.Wq