An Adaptive Disturbance Model for Model Predictive Control

Abstract

This paper proposes an adaptive disturbance model for the modeling and estimation of unmeasured disturbances. The residuals generated as a result of the output error of the process model, are used to model the unmeasured disturbances at the output. A two-stage recursive least-squares (TSRLS) algorithm is derived for the online estimation of the disturbance model parameters. The proposed scheme is capable of estimating both the stationary and non-stationary disturbances. The adaptive disturbance model is used in a model predictive control (MPC) technique to provide improved disturbance rejection. The future behavior of the modeled disturbances is predicted and incorporated into the model predictions to improve their accuracy. The effectiveness of the proposed approach is shown by the simulation example of a glasshouse process.