A Neural Network-Based Scheme for Inrush Current Diagnosis of Power Transformer

The main reason why the correct operation accuracy of transformer relay protection devices is significantly lower than that of system relay protection is due to differential protection misoperation caused by excitation inrush current. Reliable identification of excitation inrush current is the key to improve transformer protection performance. This article combines the electrical waveform characteristics of transformers under various operating states, and uses convolutional neural networks to identify the characteristics and changes of transformer voltage, differential current, and current waveforms, achieving reliable identification of excitation inrush current. Using PSCAD simulation waveforms and on-site comtrade recording files, training and testing sets are generated to verify the performance of the inrush current recognition scheme. The results show that the proposed scheme exhibits strong generalization ability for identifying transformer excitation inrush currents under different operating conditions. The electrical quantities used in the dataset are easily obtained, and there is no need to determine which side is switching-on without load. It has strong compatibility with on-site transformer protection and has good application value.