Title:Approximate Semi-explicit and Explicit Hybrid Model Predictive Control via Simplicial Partitioning

Abstract:For multiparametric mixed-integer convex programs we propose a partitioning algorithm that generates semi-explicit and explicit guaranteed real-time implementations to within a given suboptimality tolerance. The work is particularly relevant for hybrid model predictive control, where real-time on-line implementation for fast systems is hampered by the worst-case exponential runtime complexity of mixed-integer solvers. The output of our algorithm is a binary tree simplicial partition which defines a static map from the current parameter to a suboptimal solution. Convergence of our algorithm is proved by defining a novel optimal cost overlap metric, which is linked to the convergence speed. A satellite position control example shows that our approach is faster than implicit MPC by up to two orders of magnitude, allowing a guaranteed real-time on-line implementation.