Improving the Performance of Model Predictive Control by Detecting Model Plant Mismatches

The mismatch must be detected and adjusted automatically in real time to keep the performance in acceptable range and save the plant maintenance cost. Objective in this paper is to design robust controller which will detect the mismatches without requiring the maintenance of complete plant. In order to detect the mismatch, techniques of conventional correlation, partial correlation between model residuals and manipulated variables using method of least squares have been used. The comparison of Output Error (OE), Auto Regressive eXogenous variable (ARX) and Box-Jenkins (BJ) techniques concluded that Model Predictive Control (MPC) control performance can be improved using OE estimation models in the presence of mismatch.