Study on a New Method of Transformer Differential Protection

As one of the differential protection problems, misinterpretation of inrush current has caused many faults. Aiming at this problem, a new method to identify inrush current based on wavelet neural network is put forward. Firstly, this paper makes a quantitative analysis of signal-phase inrush current. After using PSCAD/EMTDC software to establish simulation models of both short circuit current and inrush current, db5 wavelet is selected to extract characteristic quantity of currents. According to the 'higher energy' characteristic of inrush current's waveform, this paper takes energy characteristic values as feature space of LM improved algorithm pattern recognition, uses the classificatory function of BP neural network to identify inrush current, and chooses 972 groups of samples to train neural network. Error distributions of all samples' output prove that this method is correct. Finally, this paper proposes a new criterion to identify inrush current. It verifies that the speed, sensitivity and reliability of transformer protection are improved by this method.