Real-Time Optimization with Modifier Adaptation Based on Dynamic Gradient Estimation

Abstract

In this paper, an optimization strategy with modifier adaptation based on a mechanistic model and the estimation of the input-output map gradient is presented to solve optimization problems in real-time. The strategy consists of defining an optimization problem based on the process model with first-order modifiers in the objective function that include the information of the input-output map gradient to correct the uncertainties of the process model. The gradient of the input-output map is estimated by means of a differentiator based on a multivariable Super-Twisting algorithm. The proposed real-time optimization with modifier adaptation strategy is implemented in a bioethanol production process and validated through closed-loop simulations.