Abstract

The control of systems composed of complex dynamically interacting sub-systems has received renewed attention recently due to important emerging application areas such as smart grids. Distributing Model Predictive Control (MPC) approaches, which are ideally suited to tackle constraints, is an active area of research with many theoretical and practical challenges to be overcome. One key element in this area is to manage the trade-off between the benefits of adding coupling between sub-systems in the problem formulation, as would be done in standard centralised approaches, and the computational downsides that this brings.

This talk will introduce a novel approach for synthesizing distributed invariant sets, restoring some of the benefits of centralized MPC forumaltions, while maintaining the sparse structure of the distributed optimization problem. The resulting predictive controllers can be synthesized in a distributed fashion, without recourse to any centralized optimization; an important requirement if the technique is to scale. We will also introduce conditions under which new sub-systems can join, or leave the network without a controller re-design, i.e., in a plug-and-play fashion, while maintaining stability and constraint satisfaction.