As the hub of railway transportation and the carrier of staff scheduling, railway station is the key factor which affects the railway passenger transport. Passenger flow has a directly effect on station operation and load redundancy, while an effective prediction of passenger flow can guarantee the station security, staff assignment and resource allocation. In this paper, the neural network is taken to predict the passenger flow, and the factors which affect the variety of passenger flow are analysis and selected as the input of neural network. The online learning and variable step rate update are adopted to estimate the number of the passenger during a certain period, which is then weighted with the historical data to derive the prediction of passenger of the next period. The simulation demonstrates that the prediction model can obtain a well result.