This paper proposes an inverter control algorithm based on the deep deterministic policy gradient (DDPG) with adjustable learning rates. This algorithm retains the advantages of model-free control of the inverter and solves the inherent problem of learning rate setting in deep neural networks. As a result, the control algorithm can quickly converge to the desired state during training, accelerate the training process, and not be affected by the control effect. Through simulation, the training results of the adjustable learning rate DDPG algorithm are compared with those of the constant small learning rate and constant large learning rate DDPG algorithms. The simulation results show that the training efficiency is improved by 75%, while the control effect on the inverter is almost unaffected.