Reconfigurable Model Predictive Control for Large Scale Distributed Systems

For large scale distributed systems, centralized model predictive control (MPC) often requires high computational resources, while generally distributed MPC can only achieve suboptimal control performance. To address these limitations, this article proposes a new reconfigurable MPC framework for large scale distributed systems, in which an optimal control problem with a time-varying structure is formulated and solved for each control loop. More specifically, at each time step, a subset of the control inputs is dynamically selected to be optimized by MPC, while the previous optimal solution is applied to the remaining control inputs. A theoretical upper bound on the performance loss, due to the fact that only a subset of inputs is optimized, is then derived to guarantee the worst-case performance. To minimize the performance loss, this upper bound is then used to guide the reconfiguration of MPC, i.e., the selection of control inputs for optimization. The applicability of the proposed approach is illustrated through case studies, including battery cell-to-cell balancing control and multivehicle formation control. Numerical results confirm that the proposed approach can achieve better control performance than distributed MPC and requires less computation time than conventional centralized MPC.