Model predictive control of large-scale systems: Application to the Tennessee eastman process

Highly nonlinear and open-loop unstable processes pose serious difficulties to the implementation of optimal control solutions, such as Model Predictive Control (MPC). An example of such processes is the Tennessee Eastman model. Here we show that by proper combination of the optimization algorithm with additional intermediate layers of control, most of the ill-conditioning can be avoided, without loss of performance. Simulation results of our control strategy are compared with previously published results. The numerical efficiency of the solution procedure is addressed, and some characteristics of the problem that can be further exploited are also identified.