Data-Driven Model-Free Predictive Control for Zero-Sequence Circulating Current Suppression in Parallel NPC Converters

This paper proposes a data-driven model-free robust predictive control strategy for parallel three-level NPC inverters based on finite control set model predictive control (FCS-MPC), focusing on the zero-sequence circulating current (ZSCC) problem under parameter mismatch conditions. A set of virtual voltage vectors with zero average common-mode voltage (CMV) is introduced to effectively suppress ZSCC without adding additional constraints to the cost function. Meanwhile, an Integral Sliding Mode Observer (ISMO) is integrated into the predictive control framework to enhance robustness and enable reliable control using only input–output data. Unlike existing studies that primarily consider ZSCC suppression under an ideal system, this work specifically addresses the practical scenario in which system parameters deviate from their nominal values. Even when ZSCC suppression strategies are employed, parameter mismatch can still lead to noticeable circulating currents, motivating the need for a more robust solution. Simulation and experimental results validate that the proposed approach achieves excellent current tracking, neutral-point voltage balance, and effective ZSCC suppression under parameter variations, demonstrating strong robustness and feasibility for practical applications.