Performance Investigation for Constrained Model Predictive Control of the Shell Fractionater Problem

Model predictive control (MPC) is one of the most profitable present day control strategies because of features like constraint handling, integrated multivariable control, optimal and safe performance. The model used to predict the future outputs in MPC plays an important part in the control calculations, thus the performance of MPC depends on the accuracy of the model used.

In this work, performance degradation is evaluated for different plant-model mismatch cases using performance assessment techniques described in the literature. Closed loop simulation in Matlab was used. A constrained MPC was formulated using a state-space model without state augmentation. The controller was used for the plant-model mismatch cases described in the Shell Oil fractionator problem using random walk signals as input disturbances. A performance index for the constrained MPC system was calculated and the results for the different mismatch cases in the Shell Oil fractionator problem were explored.