Pseudo-Three-Layer Sequential Model-Free Predictive Control with Neural-Network Observer for Parallel TType Three-Level Converters

Parallel T-type three-level converter (3LT 2 C) system has been widely concerned with its high-quality output current and improved efficiency in low-voltage applications. However, for parallel 3LT2 C, the power quality of the grid current should be considered, and the neutral-point (NP) voltage and zero-sequence circulating current (ZSCC) should be suppressed. In addition, the filter inductor mismatch can also seriously affect the overall performance. Based on the above considerations, this paper proposes a novel sequential model-free predictive control method based on an ultra-local model (ULM) and a neural-network observer (NNO) for parallel-3LT2 C. First, a cost function-free NP voltage control is designed, which significantly suppresses the NP voltage fluctuations and does not need to know the values of DCbus capacitances. Second, the ULMs of the grid-side current and ZSCC are designed, and the uncertain terms of the ULM are estimated with high accuracy by the NNO. Finally, a pseudo-threelayer sequential model predictive control is designed to simplify the weight factor selection. Comparative simulations and experiments verify the excellent performance of the proposed algorithm under parameter mismatch and different current references.