A Computation Governor for ADMM-Based MPC with Constraint Satisfaction and Setpoint Tracking

This is the authors' abstract. We don't add key points for this paper.

Model Predictive Control (MPC) exploits the numerical solution of an Optimal Control Problem (OCP). In this paper, we consider the application of Alternating Direction Method of Multipliers (ADMM) to solving quadratic programs for linear-quadratic MPC. The ADMM algorithm may only achieve primal feasibility at convergence in the limit of an infinite number of iterations. Considering this, two strategies are developed which facilitate use of ADMM when only an inexact solution can be computed with a limited number of iterations per time step. These strategies involve constraint tightening and a computational governor which modifies the reference command in the MPC problem. Simulation examples are reported that illustrate the benefits of the proposed approaches.