Abstract: The wideband and non-stationary radiated noise generated by the operation of power plant steam turbines has a continuous impact on equipment life and the acoustic environment of the plant area. To solve this problem, research is conducted on active control methods for radiated noise of power plant steam turbines based on deep transfer learning and adaptive filtering. Using pre trained deep networks to extract general acoustic features from source domain data, and fine-tuning with a small number of target domain samples to establish a radiation noise feature model with high generalization ability; Design a filtering minimum mean square algorithm control law that integrates leakage term and variable step size mechanism to achieve online self-tuning of parameters under changes in secondary channels; Introduce feedforward compensation pathway and construct a complete active control method. The experimental results show that the proposed method outperforms the comparative method in terms of noise reduction depth and steady-state fluctuation control, providing an effective solution for suppressing radiated noise from power plant steam turbines.