Research on Power Transformer Fault Diagnosis Based on Improved Wavelet Packet Energy and Hidden Markov Model

In order to realize non-invasive fault detection of power transformer, an acoustic signal analysis method based on improved wavelet packet energy transform and hidden Markov model (HMM) is proposed to diagnose the fault situation of power transformer. Firstly, the principle of acoustic vibration and the characteristics of acoustic signal of power transformer are analyzed, and the feasibility of fault diagnosis through acoustic signal characteristic component analysis is proved. Secondly, through the improved wavelet packet energy transformation method, the acoustic signal after noise reduction is effectively extracted, and the feature vector of energy distribution percentage of each running state on different frequency band nodes is constructed. Finally, the generated feature vectors are input into the trained HMM for fault diagnosis. The simulation results show that this method can effectively evaluate the running state of power transformer and provide practical basis for the establishment of fault diagnosis system.