Data-Driven Robust Designs of Performance Prediction and Its Application in High-speed Trains

This paper presents a robust performance prediction scheme for traction systems in high-speed trains. Performance variables are latent indicators of the system running, which play an important role in automated monitoring and fault diagnosis. It can be inferred from manifest variables (measurements) instead of directly observed from traction systems. Recently, most studies explored the static method for predictions, which is not suitable for dynamic systems. Considering the tricky challenges in practice, the existing prediction methods need to be further improved. Motivated by this, the paper designs a data-driven prediction model for performance variables of traction systems in high-speed trains. Specifically, to remove the influence of disturbances or noises, a robust subspace identification skill is adopted, which is developed for the soft sensor. In addition, this study further reduces the process errors of estimations in the state-space model. This method is verified on a traction system platform and an actual traction motor experiment. The experimental results prove the effectiveness and superiority of the proposed scheme.