Transformer Winding Fault Diagnosis Method Based on Improved BP Neural Network

Aiming at the problem of insufficient samples of transformer winding faults in the field and the difficulty of direct online frequency response testing, we constructed an online monitoring platform based on station transformer winding injection. By simulating short-circuit and spacing change faults between pancakes, the amplitude-frequency response curve of the response current signal is acquired and plotted using a high-frequency response current signal acquisition device. Then, statistical features and image color features are extracted from the amplitude-frequency response curve. Based on the station transformer injection method, the five most important feature parameters are selected using the random forest feature selection method. Finally, the BP neural network classification algorithm is improved by combining the dung beetle algorithm in order to realize the accurate identification and diagnosis of winding fault types.