Performance Comparision of Kalman Filter and RLS Models in MPC Applications

This paper considers a comparison between the Kalman Filter (KF) and the Recursive Least Squares (RLS) models in the design of Model Predictive Control (MPC). The assessment is validated through two industrial applications; a two-coupled tank and binary distillation column systems. The study has conducted several simulation scenarios using Matlab/Simulink software. Linear models are identified for the controlled systems where the input and output variables are subjected to non-stationary measurement of noise. In the first application, the Extended Kalman Filter (EKF) is utilized to provide a state space suboptimal model for the two-coupled tank system. The EKF is suggested as this process is characterized as a nonlinear system. The estimated model is then incorporated with the MPC control law to drive the process outputs to follow their set point trajectories. Similarly, the RLS algorithm was exploited to identify the parameters of multi-input single-output relationships for the aforementioned system. The similar policy was also implemented for the second application, which undertakes the binary distillation column. This system is classified as a linear process in this research, therefore the standard KF is utilized with application. The performance of the designed MPC scenarios, based on the two identification approaches, are evaluated by means of the Integral Squares Error (ISE) performance index.