Linguistic specifications and optimization using surrogates in computer-aided control system design

Abstract

In computational ly expensive engineering design projects it is common to use approximations as surrogates for expensive cost functions. For this purpose, the level of detail is reduced. Thus the computational cost during optimization phases becomes smaller. The essential observation in this contribution is that inexpensive surrogate cost functions may be used to accelerate (certain) direct search methods to determine a solution without sacrificing convergence, i.e. the optimization process converges to a solution to the original problem. This contribution discusses a sequence of approximation s for the cost function, and the use of such approximations as surrogates in the optimization process in a manner such that, as a result, one obtains a minimizer of the expensive cost function subject to the restrictions.