Using Parallel Computing Insights to Design and Analyze Control Architectures

Sunday, April 26, 2020 - 11:00am - 11:30am
Keller 3-180
Victor Zavala (University of Wisconsin, Madison)
Hierarchies, computing latencies, communication latencies, load balancing, and asynchronicity are issues that guide the design of both parallel computing and control architectures. This makes sense because, after all, a control architecture is nothing but a parallel computer. In this work, we show how classical parallel computing paradigms such as multigrid and overlapping Schwarz can be used to design and analyze new and scalable model predictive control (MPC) architectures. Specifically, we show that multigrid provides a general framework to design hierarchical MPC controllers and we show that overlapping Schwarz provides a general framework that spans decentralized and centralized MPC. We provide a graph-theoretical framework to unify these concepts and show that this reveals deep connections between controllability and disturbance propagation in spatial and temporal domains. Applications in energy systems are discussed to illustrate the developments. This is joint work with Sungho Shin, Mihai Anitescu, Timm Faulwasser, and Mario Zanon.