An Optimal Nonlinear Controller Design on Multilevel Boost Converter Using Reinforcement Learning

This paper presents a digital‐twin‐assisted control strategy for multilevel boost converters (MBCs) supplying constant‐power loads (CPLs), which are prone to instability due to their negative incremental impedance. A modified integral non‐singular terminal sliding mode controller (MINTSMC) is combined with a deep deterministic policy gradient (DDPG) agent to enhance transient performance while preserving robustness. The digital twin enables seamless software‐ and hardware‐in‐the‐loop (SIL/HIL) calibration, ensuring consistent controller behaviour prior to deployment. Experimental HIL results demonstrate that, while all evaluated controllers satisfy the steady‐state voltage band, the proposed MINTSMC–DDPG achieves the lowest root mean square (RMS) voltage deviation, the fastest settling time, and the smoothest control response under dynamic CPL conditions. These findings confirm that the control design suits the need for demanding power‐electronic applications such as electric vehicle charging and renewable energy systems.