A Parallelizable Interior Point Method for Two-Stage Robust MPC

Artikel i vetenskaplig tidskrift, 2017

This paper presents a parallelizable algorithm for deploying a primal-dual interior point method on two-stage model predictive control (MPC) problems. The proposed method exploits the specific structure of the problem in order to achieve a parallelizable linear algebra. The focus is set on minimizing the amount of matrix factorizations performed in order to obtain a method of low computational complexity. For commonly used benchmark problems, considering robustness against 2-50 state space models, we show that if the overhead of the parallelization is negligible, the proposed method has a computational complexity per iteration only 5%-15% higher than the state-of-the art methods for standard MPC, provided that sufficiently many CPUs are available.