Aiming at the problem of on-line detection of transformer faults

a non-invasive acoustic method is proposed to detect transformers. There are many types of transformers

and the test results are complex. Auditory alone cannot effectively detect abnormalities in transformers. A solution is proposed to solve the problems of large-scale

high dimension

and high-real-time requirements of transformer acoustic signal data. Using the collected acoustic signals of a transformer in a substation in Beijing during normal operation and abnormal operation is denoised. Using the Mel frequency cepstrum coefficient method

in the unsupervised and semi-supervised learning mode

the principal component analysis（PCA） method is used to detect the fault of the transformer acoustic signal and the effect of the PCA method on the transformer fault monitoring in the two modes is compared. The experimental results show that the transformer acoustic anomaly detection based on the principal component analysis method has a good detection ability

and in the semi-supervised learning mode

the principal component analysis method has a better detection effect.