Model Predictive Control of Linear Systems with Unknown Parameters

Abstract

This paper studies the model predictive control problem (MPC) of linear systems with unknown parameters both in system models and measurement models. The method that combines the estimation of system parameters and states with MPC is proposed, where the reinforcement learning (RL) is used to learn the optimal control strategies. Its characteristics are that the control and estimate can proceed simultaneously. Simulation studies verify that the designed algorithm can converge to the optimal linear feedback and the parameters converge as well.