Scenario Based Implicit Dual Model Predictive Control

Abstract

Model Predictive Control (MPC) has become a popular control strategy, especially within process control, where the handling of constraints and multiple inputs and outputs are essential. As the performance of the controller is crucially linked to the models predictive capabilities, model uncertainty can reduce the performance significantly. Robust MPC has been proposed to handle model uncertainty, but often leads to overly conservative solutions. In this paper, we propose a new stochastic scenario based formulation for robust MPC, where feedback is explicitly introduced in the optimization problem, to allow both state and parameter updates. The updates are conducted based on measurements from the different scenarios, and we use an Ensemble Kalman Filter (EnKF) for state and parameter updating. The resulting controller is an implicit dual MPC, and as shown in an example, applies perturbations for identification only if it will return itself over the prediction horizon.