Power Transformer is the critical equipment in electric power system.Transformer status monitor is very important for electric power system stability.This paper proposed a Transformer neural network based power Transformer monitoring system.Transformer neural network deeply mined the relationship between different input feature dimension,which offers a more stable and high accuracy power for Transformer decision system.Training data set was labeled as health status,sub health status and failure status.In feature extraction step,this paper used multiple feature fusion method which included origin data 、wavelet transform feature and Fourier transform feature.For training, we used data generator and Focal loss method to avoid data imbalance influence.Then the processed data was used for model training,we used the trainedmodel for current power transformer status prediction.Theaccuracy of the proposed method is better than traditional machine learning methods and other deep learning model,which achieve almost ninety percent.The experiment result shows this method can be used for power transformer monitor system.