Abstract: Real-time monitoring of the electric traction drive system operating temperature is essential to ensure stable train operation. To eliminate the effects of temperature measurement circuit faults including sensors and signal processing circuits, a data-driven multi-circuit temperature estimation is proposed, fault detection and fault-tolerant measurement strategy. Firstly, based on the hardware structure of the system temperature measurement and making full use of the data redundancy relationship between the four water temperature measurement circuits of the traction converter and the two oil temperature measurement circuits of the traction transformer, a temperature estimation method based on the least squares support vector machine is proposed. Secondly, a fault detection strategy for the temperature measurement circuit is proposed through the relationship between the residual distribution of the estimated temperature and the actual measured temperature of each circuit. Considering the existence of minimal measurement noise interference, a fault isolation and fault tolerance control method based on sliding probability windows is proposed. Finally, a field test was conducted. Compared with the traditional method, the proposed method has better measurement and diagnostic performance, and has a good application prospect.