Self-Training Intelligent Predictive Control for Grid-Tied Transformerless Multilevel Converters
Mohammad Babaie, Kamal Al‐Haddad
Abstract
Literature reviews affirm the constructive influence of the intelligent predictive multiobjective control (IPMOC) on the power quality, power ancillary services, efficiency, and reliability expected from grid-tied transformerless multilevel converters. Regarding, this article contributes to developing the IPMOC concept in terms of training, autonomous power management, and real-time harmonic mitigation. In the proposed IPMOC, model predictive control (MPC) adjusts the power, regulates the dc-link capacitors’ voltages, reduces the switching transitions, suppresses CMV, and alleviates harmonics using a novel real-time selective-predictive harmonic mitigation objective. As the intelligent part, two artificial neural networks trained by a novel data-free, fast, self-training strategy reinforce the MPC to handle the multiobjective task. The first one adapts the weighting factors of the MPC dynamically, while the second one as a model-free controller autonomously adjusts the active and reactive power references of the MPC to assure the unity power factor of the grid in the presence of unknown loads. The proposed IPMOC has been applied to a three-phase neutral point clamped converter and evaluated via various test scenarios run by dSPACE 1202 and MATLAB to verify its effectiveness and feasibility.