TR2018-118

An Alternating Direction Method of Multipliers Algorithm for Symmetric MPC


    •  Danielson, C., "An Alternating Direction Method of Multipliers Algorithm for Symmetric MPC", IFAC Conference on Nonlinear Model Predictive Control (NMPC), DOI: 10.1016/​j.ifacol.2018.11.051, August 2018, vol. 51, pp. 319-324.
      BibTeX TR2018-118 PDF
      • @inproceedings{Danielson2018aug,
      • author = {Danielson, Claus},
      • title = {An Alternating Direction Method of Multipliers Algorithm for Symmetric MPC},
      • booktitle = {IFAC Conference on Nonlinear Model Predictive Control (NMPC)},
      • year = 2018,
      • volume = 51,
      • number = 20,
      • pages = {319--324},
      • month = aug,
      • publisher = {Elsevier},
      • doi = {10.1016/j.ifacol.2018.11.051},
      • url = {https://www.merl.com/publications/TR2018-118}
      • }
  • Research Areas:

    Control, Dynamical Systems

Abstract:

This paper presents an alternating-direction method of multipliers (admm) algorithm for solving large-scale symmetric model predictive control (mpc) problems in real-time on embedded computers with limited computational and memory resources. Symmetry was used to find transformations of the states, inputs, and constraints of the mpc problem that decompose the dynamics and cost. We prove a key-property of the symmetric group that allows us to efficiently transform between the original and decomposed symmetric domains. This allows us to solve different sub-problems of a baseline admm algorithm in different domains where the computations are less expensive. This reduces the computational cost of each iteration from quadratic in problem size to linear. In addition, we show that our admm algorithm requires a constant amount of memory regardless of the problem size. We demonstrate our algorithm for a battery balancing problem which results in a reduction of computation-times from hours to seconds and a reduction in memory from hundreds of megabytes to tens of kilobytes.

 

  • Related News & Events

    •  NEWS   MERL Control and Dynamical Systems Group presented 8 papers at IFAC NMPC conference
      Date: August 19, 2018 - August 22, 2018
      Where: IFAC NMPC, Madison, WI
      MERL Contacts: Stefano Di Cairano; Rien Quirynen
      Research Area: Control
      Brief
      • The 6th IFAC Conference on Nonlinear Model Predictive Control (NMPC), http://www.nmpc2018.org/, is a highly focused conference that attracts experts in this area from around the world. Members of the Control and Dynamical Systems group presented 8 papers (out of the 149 at the conference!) Stefano Di Cairano delivered one of the 7 plenary lectures entitled "Contract-Based Design of Control Architectures by Model Predictive Control.".
    •