Offset-Free Model Predictive Control of the Van De Vusse Reaction

This paper presents the design of the model predictive control (MPC) of the Van de Vusse reaction (VVR) with the state estimation and compares two offset-free reference tracking techniques. The control algorithm thus compensates for the mismatches between the nonlinear plant and the linearized model, as well as other unmeasurable disturbances. In the first approach, the prediction model of the linear MPC is extended by incorporating separate state and output disturbance vectors and their heuristic smoothing. The second approach considers the augmented state-space model and fully integrates the disturbance estimation into the state estimation. State estimation is achieved using the Kalman filter, which provides optimal design in the statistical framework. The effectiveness of the proposed methodology is validated through the simulation of a nonlinear model of the VVR occurring in the continuous-flow stirred tank reactor (CSTR). Simulation results demonstrate the improvement of classical MPC with the state observer by incorporating offset-free reference tracking.