Fast identification of power transformer magnetizing inrush currents based on mathematical morphology and ANN

Discriminating between the magnetizing inrush and the internal fault of a power transform is a major challenge when designing differential transformer protection. In this paper, a novel method, which is based on mathematical morphology (MM) and artificial neural network (ANN), is proposed to solve this problem. Firstly, an MM based stage is used to extract shape features from differential currents, then after a scaling preprocessing which removes magnitude information, the features is fed into an artificial neural network (ANN) to be identified. Using this method, an effective output, which is used to block a false tripping, can be generated slightly longer than half of a period after the occurrence of an inrush condition. By using extracted features instead of raw samples of currents as inputs of an ANN, less inputs are presented to the ANN, which yields a simpler network structure. The computational load required by this method is greatly lower than those which are only based on an ANN, thus it is suitable to be implemented in resource constrained embedded systems. This method has been implemented by using MATLAB and evaluated on the data obtained from PSCAD/EMTDC simulation. The evaluation tests show promising results.