Optimal design of experiments model predictive controller

Abstract

System investigations such as simulation, diagnosis, and control require well-identified models. This work proposes an optimal design of experiments model predictive controller (MPC) to obtain experiments for identification. The main contribution is an MPC formulation with a target-oriented implementation of the parameter sensitivity (Fisher information), which remains a convex quadratic problem. Computers can optimally and efficiently solve quadratic problems, including constraints, and the method is demonstrated with a linear cathode model of a polymer electrolyte membrane fuel cell. The MPC is demonstrated in simulations, including disturbances, and significantly improves the parameter identifiability compared to a non-optimized experiment.