Distributed model predictive control approaches and distributed embedded optimization

Minh Dang Doan

Can Tho University of Technology, Vietnam, and University of Freiburg, Germany

Thursday, August 04, 2016, 11:00

Room 02-012, Georges-Köhler Allee 102, Freiburg 79110, Germany

In this talk, I present my previous work on distributed model predictive control (DMPC) and a view on extending the results to distributed embedded system controllers.

My previous work aimed at designing DMPC methods for multi-agent systems consisting of dynamically coupled subsystems, operating under coupled physical constraints. In this framework, the MPC optimization problem is convex and non-separable, and is treated using dual decomposition approaches that lead to distributed and hierarchical schemes with guaranteed feasibility, stability, and quantifiable suboptimality. Several methods additionally guarantee that the DMPC performance converges to, or even is as good as the performance of the corresponding centralized MPC.

One DMPC method that produces the same result as the centralized MPC is obtained with the idea of decomposing the computation of matrices in an accelerated proximal gradient algorithm [1]. This algorithm belongs to the class of fast first-order methods with a suboptimality of O(1/k^2), where k is the iteration number. My plan is to implement this method to embedded controllers for future tasks of distributed moving-horizon estimation and distributed MPC of large-scale infrastructure systems, e.g. estimation and control of water in an open water network.

[1]. P Giselsson, MD Doan, T Keviczky, B De Schutter, A Rantzer. "Accelerated gradient methods and dual decomposition in distributed model predictive control", Automatica, vol. 49, no. 3, March 2013.