Reinforcement Learning-Based Control for Current Regulation and Capacitor Voltage Balancing in a Four-Level Single Flying Capacitor Converter

Multilevel converters are widely used in mediumand high-power applications due to their superior features including reduced harmonic distortion and improved power quality. Among various topologies, reduced-component multilevel converters, such as the four-level single flying capacitor (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$4 ~\mathrm{L}$</tex>- SFC) converter, offer benefits in terms of hardware cost and system volume. However, a key challenge in this converter is maintaining the capacitor voltage balance due to the lack of redundant switching states. This paper proposes a Deep Q-Network (DQN)-based control strategy for simultaneous current regulation and capacitor voltage balancing in the 4L-SFC converter. The proposed reinforcement learning (RL) controller selects optimal switching actions from a discrete set of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{6 4}$</tex> possible combinations based on a state space comprising phase currents, capacitor voltages, and their respective errors. A reward function guides the agent to minimize control deviations, enabling robust performance without reliance on an accurate system model. Simulation and experimental results validate the effectiveness of the proposed method, demonstrating its potential to address the inherent control challenges of reduced-component multilevel converters.