Abstract: To address the issue that traditional adaptive frequency regulation methods struggle to adjust control parameters in real time according to dynamic system changes, thereby limiting the enhancement of frequency modulation performance, an adaptive frequency regulation control method for wind farms was proposed, based on the twin delayed DDPG (TD3) deep reinforcement learning algorithm. According to grid frequency and wind farm operating conditions, the intelligent agent optimized and learned the frequency regulation parameters during the training process, achieving dynamic and adaptive adjustment of these parameters. Results show that compared with traditional adaptive proportional-integral (PI) frequency modulation control methods, the proposed approach significantly reduces the maximum frequency deviation under various operating conditions and load disturbances, and improves frequency response performance, validating the effectiveness of the method.