Deep Deterministic Policy Gradient Algorithm Based Reinforcement Learning Controller for Single-Inductor Multiple-Output DC-DC Converter

• Published in: IEEE transactions on power electronics 2024-04, Vol.39 (4), p.4078-4090
• Main Authors: Ye, Jian, Guo, Huanyu, Wang, Benfei, Zhang, Xinan
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Control methods; Control systems; Controllers; Cross-regulation; DC-DC power converters; Dynamic response; Inductors; Interference; Machine learning; Multivariable control; Nonlinear control; Nonlinear systems; Regulation; reinforcement learning (RL) controller; single-input multiple-output (SIMO) dc–dc converters; Steady state; System effectiveness; Training; Voltage control; Voltage converters (DC to DC)
• ISSN: 0885-8993; 1941-0107
• DOI: 10.1109/TPEL.2024.3350181

Due to its advantages, such as simple structure, high power density, and strong scalability, single-input multiple-output (SIMO) dc-dc converters exhibit vast application prospects. However, a significant challenge arises in the control design of SIMO converters. This challenge stems from the shared usage of the inductor in the main circuit by all loads, inevitably leading to mutual interference and cross-regulation among the output voltages. Consequently, research on control methods for SIMO dc-dc converters has been a widely discussed topic. In this article, we propose a deep deterministic policy gradient algorithm based reinforcement learning (RL) controller. The proposed control method is applicable to multivariable and nonlinear control systems, offering excellent regulation performance and robustness. It effectively addresses the issues of mutual interference and cross-regulation among the outputs of SIMO dc-dc converters. This approach also enhances the steady-state performance and dynamic regulation capabilities of the converter when operating in continuous conduction mode. Simulations and experiments are conducted to validate the effectiveness of the proposed RL multivariable controller. The results demonstrate that SIMO converters, when controlled by the RL multivariable controller, exhibit improved steady-state and dynamic response performance. This approach effectively mitigates the cross-regulation issues among the outputs of SIMO converters.