Voiceprint State Identification of Power Transformers Based on Probabilistic Neural Network

The main equipment such as transformer and reactor of substation radiates sound signal to the outside world when it is running. The amplitude, time domain waveform and spectrum characteristics of sound are closely related to its operating parameters, mechanical state, excitation state and insulation state, which can reflect the change of equipment operating state in real time. Therefore, sound monitoring and other state quantity monitoring technology can complement each other, and comprehensive analysis of equipment state from multiple physical quantity dimensions. And the Probabilistic Neural Networks (PNN) would be used for training and recognition based on EMD and autocorrelation function. The results of the original feature and fusion feature recognition were compared and analyzed. The results show that the PNN recognition result based on the original feature parameters is only 87%, while the PNN recognition accuracy based on the fusion feature parameters can reach 93%.