Using surrogates to reduce time expenditure for optimization in systems and control

Many engineering applications of standard optimization methods result in heavy computational loads because involved objective function calculations are costly Therefore traditional optimization techniques often may become computationally unattractive or even unacceptable. An alternative to unacceptably high computational loads in optimization may be the use of objective function approximations whose calculation (and optimization) is less expensive. Such approximations are often called "surrogates", which result from simplification in functions themselves and/or in the underlying system models. From a systems and control perspective this contribution discusses threes types of surrogate functions and surrogate usage schedules (or schemes). The reduction in the number of "full" objective function calculations is sought. In such manner, using surrogate functions, constrained optimizer convergence is obtained with considerable reduction in computation time needed to determine a problem solution. Numerical results are presented which show that the proposed tools may be used to efficiently reduce computation time without sacrificing convergence.