Abstract: Because there is a big deviation between the newest intra-day predicted values and day-ahead generation schedule, by considering the characteristics that the predicting accuracy of wind power can be increased by the level with different time scales, an online rolling dispatch strategy for spinning reserve based on power deviation optimization is proposed for accommodating large-scale wind power effectively, and a mathematic model for online rolling dispatch is also built. Firstly, the combination of normal distribution with Laplace distribution is taken to simulate the forecasting error for probability distribution of wind power, then the top and bottom limitation of wind power is determined based on confidence level. Therefore, stochastic simulation is taken to get the maximum spare capacity as the limit value of power deviationconstrains. In the end, by comparing the power deviation with the limit value, the model is solved by improved particle swarm optimization algorithm. The simulation results of a real provincial power grid system verify the effectiveness of proposed strategy and model.