Experimental and Simulation Investigation of an Adaptive Model Predictive Control Scheme: Model Parametrized by Orthonormal Basis Function

The closed-loop system's performance in synthesizing model predictive control (MPC) heavily relies on the model used for prediction. In continuously operating plants, a linear model-based MPC is designed based on the operating point's linear model during the commissioning stage. However, if the plant requires significant transitions from its normal operating point, the linear model-based MPC may not be effective. Therefore, to maintain the MPC performance under changing nominal operating conditions, the model (deterministic and stochastic components) needs to be updated to predict every sampling instant. This study focuses on designing an adaptive MPC (AMPC) scheme based on the linear model estimated from the input-output perturbation data under nominal operating conditions. The OBF-ARX (generalized orthonormal basis filters with ARX structure) parametrizes the observer's dynamic components. The proposed fixed and variable pole AMPC schemes' efficacy is demonstrated using a simulation study on a binary distillation column and experimental evaluation studies on a benchmark two-tank heater setup. The efficacy of the proposed AMPC schemes in addressing both servo and regulator problems has been demonstrated through simulation and experimental results. Specifically, these schemes have been shown to effectively track set points while simultaneously rejecting disturbances. These findings suggest that the AMPC schemes hold promise for use in a variety of applications in which precise control is required.