Receding-horizon switched linear system design: a semidefinite programming approach with distributed computation

Essick V, Raymond B.

Permalink

https://hdl.handle.net/2142/102412 Copy

Description

Title

Receding-horizon switched linear system design: a semidefinite programming approach with distributed computation

Author(s)

Essick V, Raymond B.

Issue Date

2018-10-22

Director of Research (if dissertation) or Advisor (if thesis)

Dullerud, Geir E.

Doctoral Committee Chair(s)

Dullerud, Geir E.

Committee Member(s)

• Liberzon, Daniel M.
• Salapaka, Srinivasa M.
• Voulgaris, Petros G.
• Jungers, Raphael M.

Department of Study

Mechanical Sci & Engineering

Discipline

Mechanical Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Switched linear systems
• convex control
• distributed semidefinite programming

Abstract

This dissertation presents a framework for analysis and controller synthesis problems for switched linear systems. These are multi-modal systems whose parameters vary within a finite set according to the state of a discrete time automaton; the switching signal may be unconstrained or may be drawn from a language of admissible switching signals. This model of system dynamics and discrete logic has many applications in a number of engineering contexts. A receding-horizon type approach is taken by designing controllers with access to a finite-length preview of future modes and finite memory of past modes; the length of both preview and memory are taken as design choices. The results developed here take the form of nested sequences of SDP feasibility problems. These conditions are exact in that the feasibility of any element of the sequence is sufficient to construct a suitable controller, while the existence of a suitable controller necessitates the feasibility of some element of the sequence. Considered first is the problem of controller synthesis for the stabilization of switched systems. These developments serve both as a control problem of interest and a demonstration of the methods used to solve subsequent switched control problems. Exact conditions for the existence of a controller are developed, along with converse results which rule out levels of closed-loop stability based on the infeasibility of individual SDP problems. This permits the achievable closed-loop performance level to be approximated to arbitrary accuracy. Examined next are two different performance problems: one of disturbance attenuation and one of windowed variance. For each problem, controller synthesis conditions are presented exactly in the form of SDP feasibility problems which may be optimized to determine levels of performance. In both cases, the performance level may be taken as uniform or allowed to vary based on the switching path encountered. The controller synthesis conditions presented here can grow both large and computationally intensive, but they share a common structural sparsity which may be exploited. The last part of this dissertation examines this structure and presents a distributed approach to solving such problems. This maintains the tractability of these results even at large scales, expanding the scope of systems to which these methods can be applied.

Graduation Semester

2018-12

Type of Resource

text

Permalink

http://hdl.handle.net/2142/102412

Copyright and License Information

Copyright 2018 Raymond B Essick V

Owning Collections