Passenger flow prediction is an important part of urban intelligent transportation system. In order to realize accurate prediction of passenger flow, variational mode decomposition was adopted to decompose the time series into intrinsic mode function in different time scales, the long short-term memory neural network of deep learning was used to predict, and the VMD-LSTM prediction model was proposed. Data of minnesota interstate subway passenger flow were collected to validate the model. The results show that compared with the traditional LSTM prediction model, the average absolute percentage error and the root mean square error decreases by 8.38% and 256.99% respectively after improved by VMD, the prediction accuracy and robustness of LSTM neural network are improved effectively.