Abstract: A fault warning method based on the improved whale algorithm to optimize the hyperparameters of Transformer network (IWOA-Transformer) was proposed. The method improved the whale optimization algorithm (WOA) by utilizing nonlinear convergence coefficients and Gaussian variation to improve its convergence speed and avoided falling into local optimum. Then, the hyperparameters of Transformer were optimized with the improved whale optimization algorithm (IWOA) to establish a fault warning model of coal mill, and the adaptive threshold was determined by the similarity function of predicted and actual values. Combined with the expert system, the fault type was judged and solutions were proposed,and coal mill fault early warning was achieved. Finally, A fault warning test was conducted using a 350 MW cogeneration unit medium-speed coal mill as an example. Results show that the IWOA-Transformer model can significantly improve the speed and accuracy of early warning, and has practical engineering value.