The traditional model predictive current control (MPCC) heavily relies on the accuracy of mathematical models. However, there exists parameter mismatch due to modeling errors. An improved MPCC algorithm with parameter robustness is proposed to solve the problem of the controller’s poor robustness due to parameter mismatch. Through the analysis of the operating principle of MPCC, the influences of inductance mismatch on the controller are discussed. In addition, an inductance observer with division of observing area is established based on the actual and predictive current variations, thus ensuring the controller’s strong robustness. Simulations and experiments were conducted on a two-level three-phase voltage source grid-connected inverter system. In the case of parameter mismatch, the controller accurately tracked the actual inductance value and corrected the model by means of the inductance observer. Therefore, the total harmonic distortion of grid-connected current was effectively reduced. Simulation and experimental results verified the feasibility and effectiveness of the proposed algorithm.