47261-AC9
Improved Refining Operation Using Parallel MINLP Methods for Dynamic Modeling and Control of Nonlinear Hybrid Dynamic Systems
The petroleum industry operates a wide variety of chemical processes that can benefit from advanced modeling and control methods. Traditional linear methods can be applied to these systems, but this often results in sub-optimal performance of control systems. This work supported by this grant has initially considered modeling and control of a refinery simulation facility using second order Volterra series models and a nonlinear model predictive control formulation that uses hard constraints on the actual inputs. Realistic process data was generated using a high fidelity dynamic refinery model. This data was then used to determine the optimal representative model for use with nonlinear Model Predictive Control (MPC). Use of accurate nonlinear process models potentially results in improved controller performance but requires solution of a more difficult nonlinear optimization problem. Guaranteed closed-loop stability of nonlinear systems using MPC based methods generally use a terminal state constraint which may make the problem infeasible. Additionally, hard constraints on process outputs can also make the optimization problem infeasible. MPC requires the solution of an optimization problem at each sampling time. If the online optimization problem is not feasible, then some constraints should be removed or relaxed. Determining the best constraints to relax while still meeting product quality objectives can be quite difficult. This type of formulation requires incorporation of binary variables representing the satisfaction or failure to achieve a control objective. The resulting mixed-integer nonconvex nonlinear optimization problem can be quite difficult to solve, especially when attempting to solve the problem in real-time for control. As an initial approach, a method is proposed which uses iterative solution of soft constraint formulations based on a list of prioritized control objectives with inclusion of hard constraints at each iteration to ensure that high priority objectives are met. Global solution methods can also be used to provide deterministic bounds on the optimization problem as well as alternative initial starting points for the optimization algorithm. Preliminary results show that a second order Volterra model can be used to represent the multivariable chemical plant with strong nonlinear dynamics. The proposed nonlinear MPC formulation tracks set-points and rejects disturbances better than traditional linear control methods. The nonsquare formulation uses propositional logic constraints to force specific control moves that are intended to move the process state away from undesired regions thus successfully rejecting process disturbances while maintaining the highest level of product quality possible given process limitations. This work was presented at the 2008 American Control Conference in Seattle and the 2008 Nonlinear Model Predictive Control Conference in Pavia, Italy. A paper on this work has been submitted to the Journal of Optimal Control Applications and Methods. Graduate student Timur Aliyev is currently supported under this grant.
