Control and inversion-based dynamic optimization of the Tennessee Eastman benchmark problem

Abstract:
In the last years, a new class of methods for solving optimal control problem has appeared. These use nonlinear inversion and atness to reduce the number of constraints associated to nonlinear dierential equations and the number of considered variables. Additionally, state and mixed input-state constraints can be represented by means of saturations functions in order to treat them in the usual framework of calculus of variations. These transformations are performed to reduce the overall computational burden associated to numerical solving. While these methods have been proven relevant in the eld of aerospace (e.g. reentry problems or satellite clusters), it remains unclear wether they are well suited to the control of certain classes of chemical processes. Among the reasons are the dierences in encountered nonlinearities, the structure of actuation, and most importantly, the relative lack of robustness (dependability) of available models. To evaluate the potential of these nonlinear trajectory generation methods, several typical scenarios can be considered: SISO nonlinear systems, MIMO nonlinear coupled and cascaded systems, etc. This academic work will be of particular interest for process control engineers.