A Computational Analysis of Decomposition Strategies for Model Predictive Control of Resource-Constrained Dynamic Systems

This paper presents two decomposition approaches, Bilevel Optimization and Benders Decomposition, to a model predictive control of resource-constrained dynamic systems. The proposed methods yield a distributed solution that converges to the same optimum that would be obtained by a centralized controller. In this context, it is shown that the decompositions enable the use of multi-core or distributed architectures. A level regularization method is applied to accelerate the convergence of the Benders decomposition. Computational analyses from experiments with synthetic problems and a problem regarding the charging of vehicle batteries are reported and discussed, which showed that the decomposition approaches are effective at solving the distributed problems, achieving a global optimum.