No abstract

This article proposes a decentralized model pre-1 dictive control (DMPC) algorithm without communication for 2 systems consisting of multiple identical, independent actuators 3 acting on a single central plant. The particular system design is 4 relevant for applications where modularity is paramount and for 5 highly dynamic systems, e.g., battery emulator systems controlled 6 by multiple dc-dc converter modules. Each agent, consisting of an 7 actuator and a DMPC, controls a virtually scaled version of the 8 plant to implicitly consider the effects of other agents. The set of 9 DMPCs achieves the same plant performance as a corresponding 10 centralized model predictive controller (CMPC) in unconstrained 11 operation. Also, the states of the independent agents converge 12 toward the globally optimal CMPC solution. This is obtained 13 by dividing the CMPC's objective function into local objective 14 functions related to the subsystems. The optimality and stability 15 of the DMPC in unconstrained operation are shown analytically. 16 The stability of control input-constrained operation is analyzed 17 by computing the region of attraction. Numerical studies of a 18 battery emulator system compare the performance of the DMPC 19 with the global optimum in detail. 20 Index Terms-Decentralized control, power electronics, pre-21 dictive control for linear systems. 22 I. INTRODUCTION 23 D ECENTRALIZED model predictive control (DMPC) has 24 been investigated intensively since the early 2000s [1], 25 [2], [3]. It offers the possibility to apply model predictive 26 control with its strengths to systems that cannot be con-27 trolled by a centralized instance. Two of the main issues 28 of centralized model predictive control (CMPC) are: 1) the 29 complexity of solving centralized problems grows with the 30 Manuscript