Abstract

This article presents a multiagent, collaborative control framework for cascaded H-bridge (CHB) converters, designed to distribute the computational burden and reduce control hardware requirements. The proposal adopts a hierarchical structure consisting of two long-horizon model predictive control (MPC) strategies. At the upper level, an MPC-based intercluster balancing controller computes an optimal common-mode voltage reference, enforcing explicit constraints to ensure feasible and effective energy balancing among clusters. At the lower level, local cluster balancing is achieved through a distributed MPC strategy based on consensus theory. In this architecture, each local controller solves its own constrained optimization problem using only neighbor-to-neighbor communication, enabling modularity, robustness to communication delays, and real-time implementation. The effectiveness of this solution was experimentally validated using a CHB converter composed of 12 submodules, and its performance was assessed against state-of-the-art strategies under various conditions, including communication delays, channel failures, and sudden changes in control references.