Model-Free Sequential Predictive Control for MMC With Variable Candidate Set

This article proposes a model-free sequential predictive control method for an active front-end modular multilevel converter (AFE-MMC). In this approach, a generalized ultra-local model of modular multilevel converter (MMC) including grid-side powers and three-phase circulating currents is first established. Based on this, a predictor-based neural network (PNN) is deployed to estimate the unmodeled dynamics and perturbations of the MMC, which avoids the use of model parameters, leading to improved robustness performance. Meanwhile, a sequential predictive control with variable candidate set is proposed in this design to eliminate the weighting factors in the control process, while remaining the flexible control of grid-side powers and circulating currents in MMC. Finally, neither the knowledge of mathematical models nor the identification of weighting factors is required in this proposal, resulting in enhanced performance and less tuning work under both parameters match and mismatch conditions. Simulation and experimental tests are carried out to confirm the feasibility of the proposed method.