Classification of PD Sources in HV Cables using Neural Networks and the LN-FDTD Method

W.H. Barros Jr., R.M.S. de Oliveira, C.L.S.S. Sobrinho, and R.C. Leite (Brazil)


Partial Discharges, ANNs, LN-FDTD, Coaxial Cables.


The numerical method LN-FDTD (Local Non-orthogonal Finite-Difference Time-Domain Method) is used as an alternative for the simulation and fault analysis of sources of partial discharges (PDs) in a high voltage coaxial cable model. The Artificial Neural Network (ANN) technique is used for detection and classification of the cable PD sources. The training algorithm used for training the neural network is the Marquardt-Levenberg. The training procedure is performed using a set of harmonics as input. The harmonics are obtained from the difference between the registered signals in time domain, simulated with the defects and without them.

Important Links:

Go Back