A direct search approach to optimization for nonlinear model predictive control
Abstract
Nonlinear model predictive control (NMPC) depends on performing a constrained nonlinear optimization, based on predictions of future system behavior, during a sampling interval to determine the control action to be applied to the system during the next time step. The difficulty in designing an optimization procedure to solve a constrained NMPC problem is due to the finite time horizon to which the predictive model is evaluated, the state and control actuator constraints, and sampling interval length. The resulting objective function, which is to be optimized is typically not differentiable. Although there are many commercial, shareware, and open‐source optimization packages available that can perform a nonlinear constrained optimization for most cases, there are NMPC implementations requiring embedded code or that must meet stringent timing requirements that preclude the use of off‐the‐shelf packages. In cases where the predictive model is known, such as aerodynamic or hydrodynamic systems, a direct‐search optimization algorithm can perform well enough in a real‐time environment. Direct search algorithms are simple to implement and can be made more efficient by applying differential geometric techniques to the search methodology. The typical smoothness of the equations of motion for vehicular systems allows the objective function's stationarities to be handled in a straight‐forward way. Copyright © 2013 John Wiley & Sons, Ltd.
Document Details
- Document Type
- Pub Defense Publication
- Publication Date
- Dec 27, 2013
- Source ID
- 10.1002/oca.2105
Entities
People
- James D. Gibson
Organizations
- Naval Surface Warfare Center
- Office of Naval Research