DETECTION OF ANOMALOUS BEHAVIOR AND PERFORMANCE ASSESSMENT OF PREDICTIVE CONTROLLERS

This study focuses on performance assessment of model predictive control systems. A statistical methodology based on costs functions is proposed to determine a performance index which is amenable for monitoring an MPC using a benchmarking (historical) behavior and confidence control charts. An alternative methodology based on the predictability of the error is proposed for performance monitoring the MPC. The proposed methodologies are used to evaluate a DMC used to control a chemical reactor when significant changes in the process dynamics and controller tuning are introduced.