Fast Extremum Seeking of Model Predictive Control Based on Hammerstein Model

Abstract

The use of nonlinear model such as Hammerstein model in MPC will lead necessarily to a nonlinear cost function and so that a nonconvex one. Consequently, the use of a convenient optimization method to solve the resulting nonconvex problem is required. The use of the based gradient method (BGM) requires a higher computation time. Therefore the use of this type of algorithms can't be applied for system with fast dynamic. The Nelder Mead (NM) algorithm is a deterministic optimization method that does not require derivative computation. This method is able to determine the control sequence, solution of the MPC optimization problem with a low computation burden and computation time. A comparative study between the NM algorithm and the BGM based on computation time is established. These two algorithm are implemented on a SISO and a MIMO Hammerstein model.