The difficulty of integrating wind power into the grid increases with the wind power penetration level growing continuously. The prediction error of wind farm output is relatively large， and it varies with the prediction cycle and output level; therefore， different reserve capacities are needed to balance power supply and load demand. To solve the above issues， a gradual optimization method of day-ahead and within-day dispatching schedules is presented. The wind power and predication error band are involved during the implementation of dispatching schedules， and the last dispatching schedules are adjusted according to the latest predication messages to improve the accuracy of dispatching schedules gradually. The adjustments include the output power and reserve capacities of conventional generators. The proposed model is verified on IEEE 30-bus system with a wind farm. Compared to the day-ahead dispatching model， the proposed model can reflect the impact of wind power predication error on dispatching schedules more effectively， and the reserve capacities can be adjusted accordingly.