Model Predictive Control of a Three-Phase Two-Level Four-Leg Grid-Connected Converter Based on Sphere Decoding Method

Bo Long,Tianxu Cao,Wenting Fang,Kil To Chong,Josep M. Guerrero

DOI: https://doi.org/10.1109/tpel.2020.3006432

IF: 5.967

2021-01-01

IEEE Transactions on Power Electronics

Abstract:To achieve optimal control of four-leg grid-connected converters in terms of switching frequency and current tracking, the finite-control-set model-predictive-control method (FCS-MPC) can be applied. In this method, the cost function (CF) considers tracking control of grid current, filter capacitance voltage, and converter-side current as well as switching frequency before allocating different weights in its calculations. Thus, multiobjective optimization is achieved by trying to find the optimal switching sequence that minimizes the CF. However, as the horizon length is increased, the solution search enlarges exponentially, soon requiring an exhaustive search through each of many candidates. To alleviate this computational burden, this article presents an FCS-MPC method that is based on a new node-comparison sphere decoding method (NC-SDM), thereby reformulating the CF minimization problem into an integral-least-squares problem. The proposed NC-SDM reduces the computational burden associated with longer horizons by excluding as many suboptimal solutions from the candidates as possible. It does this by continuously comparing the length of two paths corresponding to each node of the search tree, and always taking the branch with shorter length. The final length of the total path is set as the initial radius after superposing all the path lengths. As a result, the initial radius estimation is much smaller than that in the Babai method and the computational cost is further reduced. Finally, simulation and experimental results validate the feasibility and suitability of the proposed method.