As for the integrated wind farm, the optimal configuration of reactive power compensation in the boost substation is studied. The hierarchical clustering method based on dynamic time warping technology is used to describe the overall fluctuation characteristics of wind power, and representative scenarios and their probabilities can be obtained. Moreover, the local fluctuation value and its corresponding probability distribution can describe the local fluctuation characteristics of wind power. Considering economy and operation reliability of the wind farm, a bi-level planning model is built. The minimization of annual total cost is taken as the upper level planning objective function, and the dynamic reactive power optimization model of the wind power is proposed in the lower level planning model. The model is solved by the genetic algorithm and the primal-dual interior point method, and the simulation results of an actual wind farm validate the feasibility and effectiveness of the strategy. This work is supported by Science and Technology Project of State Grid Beijing Economic Research Institute (No. ZZKJ-2016-05).