Abstract
In this paper, a novel distributed model predictive control (MPC) scheme with asymmetric adaptive terminal sets is developed for the regulation of large-scale systems with a distributed structure. Similar to typical MPC schemes, a structured Lyapunov matrix and a distributed terminal controller, respecting the distributed structure of the system, are computed offline. However, in this scheme, a distributed positively invariant terminal set is computed online and updated at each time instant taking into consideration the current state of the system. In particular, we consider ellipsoidal terminal sets as they are easy to compute for large-scale systems. The size and the center of these terminal sets, together with the predicted state and input trajectories, are considered as decision variables in the online phase. Determining the terminal set center online is found to be useful specifically in the presence of asymmetric constraints. Finally, a relaxation of the resulting online optimal control problem is provided. The efficacy of the proposed scheme is illustrated in simulation by comparing it to a recent distributed MPC scheme with adaptive terminal sets.
Introduction
Model Predictive Control (MPC) is a cornerstone of modern control engineering, valued for its ability to handle complex constraints by optimizing predicted future behavior. However, applying centralized MPC to large-scale systems—such as power grids or water distribution networks—often faces computational and communication bottlenecks. Distributed MPC (DMPC) addresses this by partitioning the system into smaller, coupled subsystems.
A critical challenge in DMPC is ensuring stability and recursive feasibility. This is typically achieved through "terminal sets"—safety zones that the system must reach within a finite prediction horizon. Traditionally, these sets are computed offline, have a fixed size, and are centered at the origin. Such an approach is often overly conservative, limiting the controller's "region of attraction" and causing it to fail when the system starts far from the target or faces asymmetric physical constraints, such as unequal charging and discharging limits in batteries.
Methodology
This paper introduces a DMPC scheme that overcomes these limitations by making the terminal sets adaptive and asymmetric. The approach consists of two phases:
- Offline Phase: A structured Lyapunov matrix and a distributed terminal controller are pre-computed to provide the foundation for stability.
- Online Phase: The controller optimizes not just the control inputs, but also the size ($\alpha_i$) and the center ($c_i$) of the ellipsoidal terminal sets at each time step.
By allowing the center $c_i$ to shift away from the origin, the controller can "reach out" to capture states that would otherwise be unreachable under a fixed-center regime. Furthermore, the authors propose a "relaxed" version of the optimization problem (D-RLXD). Unlike standard methods that approximate linear constraints with more restrictive quadratic ones to fit into Linear Matrix Inequalities (LMIs), D-RLXD uses linear constraints directly. This significantly reduces conservatism and further expands the feasible region.
Key Results
The proposed methods were evaluated using a synthetic simulation of an unstable discrete-time linear time-invariant system with two coupled subsystems. The performance of three schemes was compared: D-ADAP (adaptive size, fixed center), D-ASYM (adaptive size and center), and D-RLXD (relaxed version).
The primary finding is the dramatic expansion of the feasible region. As shown in the state-space trajectories below, D-ADAP fails to find a solution for initial states far from the origin. D-ASYM succeeds in these cases by shifting the terminal set center. D-RLXD provides the most robust performance, handling the widest range of initial conditions.
Beyond feasibility, the proposed schemes ensure stable regulation to the origin. The following figure illustrates the time-domain evolution of states and control inputs for D-ASYM and D-RLXD. Both methods successfully stabilize the unstable system, with D-RLXD often achieving lower total costs due to its larger, more flexible terminal set allowing for more optimal trajectories.
Conclusion
By treating the center and size of terminal sets as online decision variables, this DMPC scheme significantly reduces the conservatism inherent in traditional fixed-set approaches. The inclusion of a relaxed formulation (D-RLXD) further enhances the controller's ability to manage large-scale systems under asymmetric constraints. Future work will focus on reducing the computational overhead of these extra decision variables and extending the framework to non-linear system dynamics.
Summary updated