This study presents a new, efficient, fast and reliable technique to discriminate internal faults from no fault conditions (inrush condition, normal, over excitation and external faults with CT saturation) in 3 phase transformers. A typical 100 MVA, 110/220 KV, ??/Y 3 phase transformer connected between a 110 KV source at the sending end and a 220 KV transmission line connected to an infinite bus power system at the receiving end are simulated using PSCAD/EMTDC software. Various types of fault and no fault conditions are simulated and the differential currents are obtained. Wavelet transformation is done on the differential current and the d1 coefficients are obtained. The d1 coefficients are given as inputs to the wavelet based neural network trained by Particle Swarm Optimization (PSO-WNN). The simulation results show that PSO-WNN has very simple architecture, negligible error and provides more accurate results when compared to wavelet combined neural network trained by back propagation algorithm (WNN) and neural network trained by giving 3 phase differential currents as input (ANN). The performance of PSO-WNN based relay is also compared with the conventional harmonic blocking relay. PSO-WNN based relaying provides a high operating sensitivity for internal faults and remains stable for no fault conditions of the power transformers.
展开▼
