Abstract: In train braking system pipeline leakage, the detection and allocation of brake main air pipeline leakage, and brake cylinder and associated pipeline leakage are extremely difficult. The data-driven detection and early alarming method of metro vehicle braking system pipeline leakage is introduced. This method obtains raw data from TCMS (train control and management system) and establishes model after processing the data. A machine learning model based on anomaly detection model and regression model is adopted to determine the health field in the space, and different parameters of the model are used to characterize the leakage at different locations. Early warning information about leakage is provided according to the variation law of different parameters over time. In view of the problem of limited sampling rate of raw data and quality of data transmission, the physical quantity definition method of ‘ratio’ and abnormal data segment elimination method are adopted. The test results show that the model can monitor and locate leakage points and provide early warning.