<p>To address the issues of high harmonic content and poor robustness in traditional model predictive current control for single-phase neutral-point-clamped three-level inverters, this paper proposes a double-vector model-free predictive current control method based on a linear extended state observer. First, the proposed method establishes an extended state observer using a first-order ultra-local model of the system, which estimates and compensates for the aggregate disturbances of the system in real time, enhancing the immunity and robustness of the system. Second, two voltage vectors are applied within each control cycle to improve the accuracy of predictive control. Additionally, the cost function is evaluated based on voltage vectors rather than switching states, which reduces the computational complexity and current harmonics. To comprehensively validate the effectiveness of the proposed method, a single-phase three-level inverter experimental platform is developed based on a DSP-TMS320F28374D chip. The effectiveness of the proposed method is demonstrated through a detailed comparison with traditional double-vector model predictive control in terms of steady-state performance, dynamic response speed, and robustness.</p>