Traction transformers face challenges, such as complex operating environments, communication interference, and sensor malfunctions, which result in missing or inconsistent data, thereby affecting the equipment condition prediction models. To address it, a Wasserstein generative adversarial imputation network with gradient penalty (WGAIN-GP) is employed to impute missing data, which is then utilized in the model prediction. Imputed data generated by various imputation methods were added as augmentation data to a Sequence-to-sequence (Seq2seq) model for predicting traction transformer status. Results indicate that the WGAIN-GP demonstrates superior imputation performance, and the imputed data significantly improve the prediction accuracy of traction transformer status.