Design of Model Predictive Controller for a Biological Fermenter

Abstract

Model Predictive Control (MPC) is a control strategy which utilizes a process model to compute a sequence of control moves with a desired control objective of tracking the desired level of the controlled variable. Extensive work has been undertaken to determine tuning strategies for model predictive controllers. However, much of that work has focused on determining the optimal tuning parameters for a particular set of process conditions, or has ignored the presence of uncertainty in the process model. This work aims to develop robust model predictive controllers by explicitly accounting for plant-model mismatch. To do this, model predictive controllers are created for three second-order process models obtained from finding the best-fit transfer function to open-loop step-response data obtained from a microbial fermenter. Further, the optimal MPC settings (namely the control horizon, prediction horizon, and the weights) are determined for the nominal case when there is no uncertainty. The optimal settings for the nominal scenario are used to inform the optimal settings for the uncertain scenario, which are found by randomly generating 500 mismatched process models based on observed experimental uncertainty. Graphical techniques are used to find the optimal settings that maximize the control performance and minimize variation in the control performance in response to step changes in the set-point and the disturbance.