Abstract: Due to the inhomogeneity of power cyber physical system（CPS） network and the dynamics of risk propagation process,it is difficult to determine the critical point of risk outbreak. From the perspective of dependent network, this paper proposes a method for determining the risk propagation threshold of power CPS network based on percolation theory. Firstly, according to topological association and coupling logic, the power CPS network is abstracted as a dual layer complex network with directed unauthorized graph. Asymmetric balls-into-bins distribution method is used to establish"one to many"and"partial coupling"CPS characterization models for non-uniform electric power. Then, considering the directionality between information layer and physical layer links, the percolation probability is introduced to establish the propagation dynamic equation for internal coupling relationship of each layer. Finally, the risk propagation threshold is solved by defining the survival function of the nodes in power CPS network, and the effectiveness of the proposed method is verified by an example of IEEE 30-bus system and 150-bus BarabsiAlbert model.