The transformer rectifier unit (TRU) is one of the key power conversion devices in the secondary power supply system of an airplane. During the operation of the TRU, it is susceptible to temperature and humidity variations and load fluctuations, leading to corresponding failures of its components, which reduces the reliability of the equipment and then affects the safety of flight. In view of the problem that TRU hardware has many fault categories and similar fault data characteristics, a fault diagnosis method based on stacked denoising auto encoder (SDAE) combined with genetic algorithm (GA) to optimize the Transformer is proposed. The following is an example of the optimization of Transformer’s fault diagnosis method. First, the collected fault data are normalized; second, the contrastive center loss (CCL) function is introduced in the training phase of SDAE to learn the optimal classification features in the layer-by-layer nonlinear mapping of SDAE by using the sample label information, so as to realize the reduction of the distance within classes and the expansion of the distance between classes. At the same time, the CCL and reconstructing cost losses (RCL) function are jointly optimized to obtain the improved SDAE-based feature extraction module, which realizes the feature pre-extraction of the original fault data; in order to further extract the feature information and diagnose the problem, the diagnostic module of the GA-optimized Transformer is constructed to improve the accuracy of fault detection. Finally, Simulink is utilized to simulate the fault data to compare with the existing diagnostic methods. The results show that the proposed method can better realize the diagnosis of 101 kinds of faults, with an accuracy rate of 96.05% and good noise resistance.