Abstract

A three-level T-type converter interfaced with an LCL filter (LCL-3LT2C) has been widely used in low-voltage and high-quality photovoltaic (PV) systems. The control of LCL-3LT2C is a multiobjective optimization problem (MOOP), which is suitable to apply model predictive control (MPC). Although the MPC has fast dynamic response, accurate transient control, and easy handling of MOOP problems, its control performance is still highly dependent on the system model used for the control system. If the system parameter mismatch between the plant and the MPC controller occurs, the performance of the control object will degrade. To solve this problem, a novel adaptive ultralocal model-free predictive control for LCL-3LT2C is proposed. First, the relationship between the NP voltage and the converter current is analyzed, and a dynamic antidisturbance factor is introduced into the reference of the converter current for NP voltage suppression. Second, to improve the model estimation accuracy, a new ultralocal model is introduced, which decouples the known and unknown terms of the system into the linear known and the nonlinear unknown terms. Finally, an extended sliding-mode disturbance observer is considered to estimate the nonlinear unknown term. The proposed method simplifies the control of NP voltage, converter current, and grid current. Experimental results on the 10-kW laboratory prototype demonstrate its superior performance over existing techniques. © 1986-2012 IEEE.