Transformer Inrush Current Recognition Based On Convolutional Neural Network

Transformer,as the energy exchange hub of the power system,undertakes the tasks of voltage transmission and conversion,and plays a vital role in the power system.Once the transformer fails,it will cause a large-scale power outage,which not only affects people’s normal lives,but also causes serious economic losses.As the main protection of transformer protection,differential protection can directly and reliably operate the transformer,which directly affects the safe and stable operation of the transformer.How to accurately identify the transformer magnetizing inrush current,the inrush current and the short-circuit fault current are the main problems of differential protection.To study a new method for identifying inrush current and inrush current,which can effectively improve the selectivity,reliability and sensitivity of transformer protection.Theoretically analyze the principle of excitation and inrush currents generated by transformers.Use MATLAB/Simulink simulation software to build simulation models of excitation and inrush currents and fault currents,and analyze the characteristics of excitation and inrush currents and fault currents.Aiming at the different characteristics of excitation inrush current,inrush current and fault current waveform,a transformer inrush current identification method based on BP neural network is proposed.The simulated excitation inrush current,inrush current and fault current waveforms are used as the original data,the statistical characteristics are extracted as the input of the BP neural network,the working principle of the BP neural network is analyzed,and the BP neural network model is designed to identify the current type.The results show that the identification method of transformer inrush current based on BP neural network has a recognition accuracy of 93%for excitation inrush current,inrush current and fault current,and can correctly identify the three currents generated by the transformer.Taking advantage of the unique advantages of convolutional neural networks in image processing,a transformer inrush current recognition method based on convolutional neural networks is proposed.This paper proposes a method for constructing a two-dimensional grayscale image.The simulated excitation inrush current,inrush current and fault current waveform data are constructed into a two-dimensional grayscale image as the input of a convolutional neural network.A three-layer convolutional neural network that can be used for transformer inrush current identification.The recognition results show that the transformer inrush current identification method based on the convolution neural network can accurately identify transformer excitation inrush current,inrush current and fault current.The recognition accuracy rate is as high as 97.5%.The method reduces the feature extraction link,simplifies the recognition process,and improves the recognition accuracy.