A novel reinforcement learning controller for the DC-DC boost converter

In this paper, a robust control for the DC/DC boost converter to regulate capacitor output voltage is studied. Boost converters are highly nonlinear systems that exhibit interconnected state variables and system parameter variations due to load changes. To cope with these characteristics, A DC/DC boost converter controller that uses the twin delayed deep deterministic policy gradient algorithm (TD3) control in continuous state is developed. This controller objective is to keep the output voltage constant under input voltage variations with fast response, little overshoot/undershoot, and ripples. As large fluctuation occurs in the actions under direct TD3 control, a new control that uses a compensatory-TD3 (C-TD3) controller is newly proposed to address this issue. This study proposes a detailed method for verifying policy networks through hardware-in-the-loop (HIL) testing and experimentation. Under three different operating conditions, the C-TD3 controller reduced settling times by 0.27s, 0.22s, and 0.18s compared to the PI controller, improving the response time of 22.5 %, 19.2 %, and 15.7 %, respectively. In addition, the integral absolute error of the C-TD3 controller was decreased by 0.279 V, 0.366 V, and 0.493 V, respectively. The results show that under C-TD3 control, voltage fluctuations were smaller, the response time was faster, and the duty cycle signal was more stable compared with those under PI control.