Aiming at the nonlinear and non-stationary characteristics of high-speed train bogie vibration signal and incomplete information caused by single-channel fault diagnosis， a method is proposed combined multivariate empirical mode decomposition （MEMD） and full vector intrinsic mode function （IMF） information entropy to extract the feature of high-speed train. Firstly， the MEMD method is used to decompose the vibration signal of homologous dual channels， and a series of 2 dimensional intrinsic mode functions are obtained. Then， the full IMF information entropy of the first 6 IMFs is calculated respectively， and the feature dimension reduction is performed by feature evaluation method. Finally， the obtained feature vector is used as an input of the support vector machine to identify the fault type of the bogie. The experimental results show that the method can effectively improve the fault recognition rate of the bogie， up to 100%， and verify the feasibility of full-vector IMF information entropy in high-speed train fault diagnosis.