Nonlinear model predictive control of the Tennessee Eastman process

This paper aims to illustrate several key issues in the implementation of a conventional-nonlinear model predictive control algorithm on a reasonably large industrial process and to test the effectiveness of the nonlinear model predictive control algorithm proposed by Zheng (1997) for control of large nonlinear systems with constraints. We show why a conventional nonlinear model predictive control algorithm may fail to provide integral control under very reasonable conditions (i.e. integral control is guaranteed if and only if a global solution is implemented and the output horizon is infinite) and illustrate this undesirable behavior through simulations on the Tennessee Eastman process. In addition to computational advantage, we argue that Zheng's algorithm may be preferred based on robust performance consideration.