Parameter Estimation for Multistage Processes: A Multiple Shooting Approach Integrated with Sensitivity Analysis

The predictive capacity of models is becoming increasingly relevant due to the increase in the development and deployment of digital twins in several areas of manufacturing. Therefore, it is important to keep the models up to date so that they represent the process reliably. One way to keep these models calibrated is via parameter estimation. However, parameter estimation problems in nonlinearly parametrized systems result in local optima or suffer from high computational costs. In view of the aforementioned limitations, multiple shooting parameter estimation arises as viable method, since it provides a setting to deal with the ill-defined regions of the search space. In this paper, we propose an improvement to the multiple shooting parameter estimation integrating sensitivity analysis. The approach divides the measured operating trajectory into different segments and performs a sensitivity analysis to find the most contributing parameters in each of these segments. In this way, the proposed technique benefits from the selection of a subset of the most sensitive parameters per segment and the computational advantages of the multiple shooting method. The performance of the multiple shooting parameter estimation integrated with sensitivity analysis is tested on a case study, a batch reactive distillation column which is a multistage process.