The work efficiency of urban rail transit mostly depends on the train schedule strategy, as it could influence the balance of energy saving and passenger flow pressure to station site. In this paper, a new approach to train dispatching and schedule is introduced to optimize metro operation. Each station site adds up the amount of coming and exiting passengers as history data, and then, passenger flow prediction is carried out through Kalman filter. After data analyzing, the center server adjusts the train timetable dynamically using the exponent function.