Abstract: To improve the intelligence level of trains and the efficiency of on-site maintenance, in this paper, a real-time diagnosis method based on fault time-series feature pattern recognition was proposed from a system perspective to solve the problem of accurate fault location on the line-side of the high-speed train traction drive system. Firstly, the system signals associated with the fault were selected based on mechanism analysis. Secondly, the correlation relationships between the fault source and the system signal were revealed and fault feature indicators were extracted by combining actual case waveform data and expert experience. Then, based on the time-series change characteristics of the fault feature index, the Gaussian mixture models and hidden Markov models (GMM-HMM) algorithm was used to establish a fault diagnosis model for the identification of the time-series feature of the over-current of line-side in traction drive system. Finally, the actual train operation data was used to verify the fault diagnosis model. The test results show that the algorithm proposed can achieve effective fault detection and isolation, and has good application value.