Reliable and Efficient Optimization Strategies for Nonlinear Model Predictive Control

Abstract

Previous experience with predictive control algorithms has shown that the way the optimization problems are formulated and solved has a big impact in the success of the control strategy. Here a multiple shooting formulation is proposed, where a process model is integrated separately inside each sampling interval, and the corresponding equality constraints are added directly to the optimization problem. It is shown that the resulting formulation provides a more reliable framework for the solution of predictive control problems, both in the linear and nonlinear cases. This strategy is compared with the original nonlinear Newton-type (state space) algorithm, on a number of process models with challenging features, including the reactor model from the Tennessee Eastman problem.