In order to adapt to active distribution network which will have large-scale distributed generation (DG) and high level of automation in the future, active management measures must be considered in the planning stage to optimize system operation. And the existing research of distribution network reactive compensation planning ignores the uncertainty of DG and load. Aiming to solve the problems, this paper proposes reactive compensation optimal allocation model in active distribution network based on the uncertainties of intermittent DG and load power. The lowest reactive compensation capacitors investment cost and network loss cost are taken as the objective of upper level. And considering two active management measures including regulating reactive power compensation capacity and regulating the on-load tap changer, the lower level optimizes for each scene. This paper uses K- means clustering method to reduce the scene, and combines with the harmonic search algorithm and particle swarm algorithm to solve the model. And simulative results of IEEE-33 distribution system verify the effectiveness of the model and solution methods. This work is supported by National High-tech R & D Program of China (863 Program) (No. 2014AA051901).