The access of large-scale wind power brings great uncertainty to the operation and control of a power system, and also brings challenges to the system restoration after blackout. Dynamically adjusting the partition restoration scheme according to the changing wind power output scenarios during the restoration process helps to improve restoration efficiency. Based on the uncertainty of wind power in the initial outage scenario, an online dynamic partition restoration optimization method for a power system is proposed to further consider the uncertainty of wind power output during the restoration process. First, the uncertain scenario set of wind power output is established, and the measure between distributions is constructed based on Wasserstein distance. Kernel density estimation is used to obtain the uncertain set of wind power output prediction error. Secondly, the restoration model, partition model and dynamic partition constraints are characterized. Two-stage optimization objectives are set up from the perspectives of grid topology and operation state, and the two-stage dynamic partition restoration distributed robust optimization model is established. The dual theory is used to realize the transformation and analysis of the model. Finally, a simulation of the New England 10-machine 39-bus system and a real power system verify that the dynamic partition restoration method proposed can effectively deal with the uncertainty of wind power output and improve system restoration efficiency.