Traditional centralized wind farm active power control strategy has a series of problems such as excessive calculation load on the central controller and too expensive construction of the communication network when the scale of the wind farm is large. The convergence of fully decentralized control is significantly affected by communication delay and the number of iterations. In order to find an active control strategy suitable for large-scale offshore wind farms, this paper combines distributed control and centralized control to transform the traditional single-layer control structure into a bi-level distributed control structure. The upper level control is designed based on a distributed consensus protocol, which reduces the construction investment in the communication system and relieves the solution burden of the centralized controller. The lower level control uses a vertical distributed control strategy based on Model Predictive Control (MPC) and Alternating Direction Multiplier of Multipliers (ADMM) to improve convergence speed. Experiments prove that the proposed method decomposes the large-scale constrained optimization problem into multiple small-scale problems that can be solved in parallel, and finally achieves an optimal control effect. This work is supported by the National Natural Science Foundation of China (No. 51877127).