The fault early warning of the coal mill is of great significance to the safe operation of thermal power unit

but the operation of the coal mill has many interference noises and a high degree of coupling

which makes the fault early warning more difficult. Based on this

this paper proposes a fault warning method based on wavelet packet transform(WPT) and Transformer. Firstly

the signal is denoised by the wavelet packet analysis method with adaptive threshold value. Then

the characteristic parameters related to the fault measurement point are selected as input to establish a Transformer coal pulverized prediction model based on the self-attention mechanism. Finally

the kernel density estimation method is used to analyze the prediction deviation and determine the warning threshold. Taking a 660 MW medium-speed coal mill as the research object and using actual data for verification

the experimental results show that the prediction accuracy of the proposed method is higher than that of CNN

LSTM

and CNN+LSTM models

and it can provide early warning of coal mill failures.