Abstract
This paper addresses the development of new formulations for estimating modeling errors or unmeasured disturbances to be used in Model Predictive Control (MPC) algorithms during open-loop prediction. Two different formulations were developed in this paper. One is used in MPC that directly utilizes linear models and the other in MPC that utilizes non-linear models. These estimation techniques were utilized to provide robust performance for MPC algorithms when the plant is open-loop unstable and under the influence of modeling error and/or unmeasured disturbances. For MPC that utilizes a non-linear model, the estimation technique is formulated as a fixed small size on-line optimization problem, while for linear MPC, the unmeasured disturbances are estimated via a proposed linear disturbance model. The disturbance model coefficients are identified on-line from historical estimates of plant-model mismatch. The effectiveness of incorporating these proposed estimation techniques into MPC is tested through simulated implementation on non-linear unstable exothermic fluidized bed reactor. Closed-loop simulations proved the capability of the proposed estimation methods to stabilize and, thereby, improve the MPC performance in such cases.