The reliability of the train system is impacted by the running state of high-speed trains. To analyze the running state, and aimed at the vibration acceleration signal of high-speed trains, a characteristic analysis method of segmentation-energy entropy and singular entropy is proposed. Firstly, the characteristics of the vibration acceleration signal changed with velocity are analyzed. Then, the signal is divided into N range. Segmentation-energy entropy and singular entropy are extracted to make feature vectors. Finally, the support vector machine method is used to classify and identify faults. Experimental results show that four kinds of typical working conditions can be accurately identified in lateral acceleration features of the car body in different speeds, that the identification ratio is above 95%, and that the fault state of high speed trains can be identified effectively.