Virtual Reference Feedback Tuning With Data Selection Criteria Applied to a Thermal Process
Keywords: Data-driven control, VRFT, data selection criteria, singular values, condition number, errors-in-variables problems, ordinary least squares, data least squares
AbstractAdjusting the controller's parameters using a data-driven (DD) methodology usually requires data gathered from a specific experiment performed in the process, which may be a time consuming task for the designer. To avoid this task, routine operating data could be used instead, however, combining those raw data along with a DD method almost certainly results in inappropriate tuning. Therefore, it is advisable to pre-select the useful information before estimating the controller's parameters. The present work is an extension of our previous works, where two data selection criteria were applied to the Virtual Reference Feedback Tuning (VRFT) method. In the present work, we have combined the application of those criteria to select the relevant data subsets. Moreover, the controller's parameters are estimated using not only VRFT's original solution, known as Ordinary Least Squares (OLS), but also the Data Least Squares (DLS) solution. The feasibility of the proposed solution is evaluated through experiments carried out in a thermal process.