To mitigate the impact of sudden sensor faults in traction rectifiers during high-speed train operation, this paper proposes a sensor fault-tolerant control method based on an improved sliding mode observer, which achieves rapid sensor fault diagnosis and fault-tolerant control. Firstly, the topological structure of a two-level traction rectifier is analyzed to establish its mathematical model with unknown disturbances. Subsequently, a novel sensor fault diagnosis strategy employing an adaptive sliding mode observer is developed. The sliding mode observer is designed using linear matrix inequality (LMI) methods, and the sliding mode robust term gain is adjusted online via adaptive laws to obtain voltage and current estimation signals. Residual signals are constructed by comparing these estimated signals with sensor-sampled data, enabling online fault diagnosis through threshold comparison. Furthermore, a voltage outer-loop sliding mode fault-tolerant controller is designed by introducing a novel sliding mode exponential reaching law, complemented by a current inner-loop fault-tolerant controller to enhance system robustness. Finally, a high-power traction rectifier model was constructed on the Simulink platform for simulation verification in this study. The simulation results demonstrate that the proposed fault-tolerant method can diagnose sensor faults within 0.020 s and implement compensation within 0.050 s, with observer state estimation error maintained below 1%. The voltage outer-loop fault-tolerant controller achieves a reduced settling time of 0.035 s under load transient conditions. The proposed methodology effectively enhances system dynamic response performance and operational reliability, providing valuable insights for safety-critical control of high-speed trains.