ADMM algorithms have been applied to a variety of problems in the last few years due in part to: the simplicity of the iteration and the ability to exploit problem structure. In this talk, we present an ADMM algorithm for the class of convex quadratic programs (QPs) that arise in the context of Model Predictive Control (MPC) applications. Our convergence analysis allows to establish the convergence rates of the algorithm for feasible and also optimal value for the ADMM parameter (or step-size) that maximizes the rate of convergence. We also present a combination of the ADMM algorithm and Conjugate Gradient method to accelerate convergence. Numerical results on the performance of the algorithm will be presented on benchmark QPs and QPs from Model Predictive Control applications. Time permitting results on infeasible QPs and extensions for 2-stage stochastic QPs will also be presented.
More details can be found at: http://arxiv.org/abs/1411.7288
Arvind Raghunathan is a Senior Principal Researcher at Mitsubishi Electric Research Laboratories (MERL). His research focuses on algorithms for optimization of large scale nonlinear and mixed integer nonlinear programs with applications in power grid, transportation systems and model-based control of processes. Prior to MERL he worked at United Technologies Research Center developing optimization algorithms for aerospace, elevator, energy systems and security applications. Arvind obtained his Ph.D in Chemical Engineering from Carnegie Mellon University and is the co-author of 30 publications and 5 patents.