GA trained parallel hidden layered ANN based differential protection of three phase power transformer

Abstract

Transformer protection is an established area of research to find the fastest and efficient differential relay algorithm that isolates the transformer from remaining system causing least damage. Algorithm should also avoid mal-operation when differentiating between the operating conditions. Various differential algorithms were proposed in the past, allowing a scope for further research. In this paper, ANN is used as a pattern classifier which discriminates among normal, magnetizing inrush, over-excitation and internal fault currents in a power transformer. The proposed scheme has been realized through different ANN architectures including a new customized parallel-hidden layered design, which originates to be more accurate in differentiating between the normal wave and faulty wave despite the shape similarity. A combination of two ANNs in Master-Slave mode has also been discussed. Back Propagation (BP) and Genetic Algorithm (GA) are used to train the multi-layered feed forward neural network and their simulated results are compared. The neural network trained by GA gives more accurate results (in terms of mean square error) than by BP Algorithm. Simulated data are used as an input to the ANN to verify the accuracy of the algorithm. Thus, GA trained Master-Slave ANN based differential protection scheme provides faster, accurate, more secured and dependable relay for power transformers. © 2014 Elsevier Ltd. All rights reserved.