It is difficult to directly measure the effective wind speed at the hub of large wind turbines. Traditional wind speed estimation methods have hysteresis. The Taylor frozen turbulence assumption ignores the changes in the wind field structure from the lidar measurement point to the hub, which affects the accuracy of the measurement data. Aiming at the above problems, the auto-regressive moving average and external input (Armax) model were used to model the wind evolution process. The particle swarm optimization algorithm is used to estimate the model parameters, and the inertia weight of the conventional particle swarm algorithm is improved to avoid falling into a local minimum. In order to ensure the real-time and fast control action of the wind power system, the effective wind speed at the hub is predicted one step in advance according to the established model. Using Fast and Matlab/Simulink software, the joint simulation is carried out with an average wind speed of 7 m/s and a turbulence level A as an example. The simulation results show that the proposed method has higher real-time performance and accuracy and is more effective than traditional method of wind speed estimation.