APPLICATION OF SOFT COMPUTING TECHNIQUES IN POWER TRANSFORMER PROTECTION

Abstract

Transformer is most important classes of hardware in electrical power system. It is one of the important links in a power transmission & distribution system. The consequence of even rare fault may be serious unless the transformer is quickly disconnected from the system. A different protection scheme applied to power transformer depending upon its rating, location in grid. A differential relaying may give false tripping in case of certain system operating condition viz. CT mismatching, CT saturation during through fault condition, inherent phase shift on primary & secondary side of the current of transformers, CT ratio error, tap changing operation, magnetic inrush current, saturation of transformer core due to over excitation, under frequency ,overvoltage operation, sympathetic inrush. Abovementioned condition always challenges the researcher to discriminate between actual internal fault & the above operation. In this report, various operating condition of power transformer and its simulation model using PSCAD has presented in details. The harmonic restraint method, used extensively, is based on fact that second and fifth harmonic component of inrush current are considerably larger than typical internal fault current. This method fails to discriminate between inrush and internal fault conditions in modem power transformer ,which are designed with new core materials , high flux density and high second harmonic generated in both condition. Most of researchers have used neural networks for differentiating between different operating conditions of power transformer. On same line, in initial portion of work, ANN based scheme is considered because major advantage of ANN viz. trainability , speed of computation, robustness, generalization, adaptation to changes in data and ability to deal with non-linear processes. Further due to their nonlinear nature, they can define nonlinear decision surface in classification problem. In this work, four different type of scheme has been developed for discrimination between internal fault and inrush operation , and transformer event monitoring , namely Back Propagation Neural Network ( BPNN ), Radial Basis Function Neural Network ( RBFNN) , Radial Basis Probabilistic Neural Network (RBPNN ) and Support Vector Machine (SVM ). ii