A strategy for controlling DC boost converters under disturbances is proposed through a disturbance-resistant reinforcement learning method. An iterative training process is designed using model-free deep reinforcement learning. A nonlinear disturbance observer is integrated to enhance the controller’s adaptability to disturbances. The proposed neural network controller achieves a maximum voltage deviation of 4.2% of the reference value, and a settling time of 6.5 ms in simulation studies. During experimental validation, the voltage deviation is limited to 6.0% with a settling time of 8 ms under the same operating condition. Simulation and experimental results demonstrate that the proposed control strategy is able to stabilize the output voltage under various disturbances, offering better performance under large-signal fluctuation. This work extends the practical deployment of reinforcement learning in the field of nonlinear control.
Yu et al. (Thu,) studied this question.