Classification of Transformer Abnormal Current Types Using Machine Learning

• Main Authors: Chieh-Chun Hsiao, 蕭傑駿
• Other Authors: Cheng-Chien Kuo
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/96f3x3

碩士 === 國立臺灣科技大學 === 電機工程系 === 107 === This thesis using machine learning to classify the abnormal current types of transformer. When the transformer connect to the power system or when the external faults is cleared, the residual magnetic flux cause the transformer core to saturate and may cause a transient electromagnetic inrush current. The inrush current is different from the external faults current, so how to classify the abnormal current types, and prevent the protection relay to malfunctioning is important. Inrush current and external faults current have different characteristics, and these characteristics are regarded as input data of neural network. The transfer function of neural network processing unit adopts tangent hyperbolic function, and is structured by the multi-layer feedforward network. In order to classify the abnormal current types of the transformer. This thesis using Matlab Simulink to simulate the inrush current and the external faults of the transformer. The simulated currents using discrete wavelet transform to capture the important data of the signals, and then the statistical features are calculated as the input data of neural network. After the neural network learning and recalling. Finally it can classify the abnormal current types of the transformer. The method proposed in this thesis is feasible, and the classification results have good accuracy.