Deep Reinforcement Learning Based Point Level Suppression Strategy in Three-Level Inverters

In three-level inverters, the mid-point level offset affects the stability of the system, the harmonic content and the quality of the output voltage. The traditional mid-point level balance control method is too dependent on an accurate system model and is sensitive to parameter changes. A control strategy based on deep reinforcement learning (DRL) with strong adaptive capabilities and nonlinear processing capabilities is proposed. The proposed intelligent control method is based on DRL three-level output voltage control and mid-point level fluctuation suppression. This method establishes an environment and an observer by simulating the control environment through the establishment of an agent and a simplified model, and collects intermediate state variables, so that the agent can achieve variable transfer and control target optimization through deep reinforcement learning, effectively reducing output harmonics, stabilizing output voltage fluctuations, and suppressing neutral point voltage fluctuations.