Effect of Model Plant Mismatch in Model Predictive Controller Performance: Continuous Stirred Tank Reactor

Plant model is one of the important aspects in the design and implementation of Model Predictive Controller (MPC). The performance of MPC depends on the accuracy and quality of plant model. However, dynamic behaviour of a plant may change with time. Hence, plant model that are used for the design will no longer represent the plant current state after some time. In this dissertation, the effect of model plant mismatch on MPC performance will be shown by the researcher. During the conduct of this research, the researcher has developed a non-linear CSTR model by using SIMULINK. Manipulated variable and controlled variable for the CSTR model has been set by the researcher. Besides that, the researcher developed 3 different linear transfer function model using 3 different ranges. By using this 3 different transfer function model, the researcher designed 3 different MPC. The researcher has tested the plant model with 2 different tests. First, to understand the dynamic model of this CSTR, the researcher has done an open loop test to this CSTR model by adding few percentages of increment in step change to the plant input. The changes in controlled variable inside the reactor is then measured and analyzed. For the second test, the researcher done a closed loop test to measure the performance of MPC between the accurate plant models and mismatch plant models. This test is done by using MPC with plant model to control to a limit which is out of its range to represent the mismatch plant model. In the open loop test, when step change is added to the plant input, all output changes from its set point which clearly shows the non-linearity behaviour of the plant. For the MPC performance test, when mismatch is added, the controller becomes less stable and it took a longer time to reach the steady state and the new set point.