The stability of wind turbine output power is seriously affected by the uncertainty of wind turbine model and external disturbances such as wind speed. The control strategy based on accurate fan parameters is difficult to meet the control requirements of the system. Therefore, a fan pitch controller based on the DDPG algorithm was proposed in this study. With the advantage that reinforcement learning only needs to interact with the environment without modeling, the fan model was taken as the training environment, the power was taken as the reward target, and the paddle angle was taken as the output. The deep neural network was used to build the actor-critic unit to train the optimal paddle variable strategy. Three typical wind conditions, namely step, low turbulence and high turbulence, were used to test the algorithm.The simulation results showed that the controller based on the DDPG algorithm had better control accuracy, overshoot and regulating time than the traditional proportional-integral-differential controller under different wind conditions.