The Use of Surrogate Modeling Algorithms to Exploit Disparities in Function Computation Time within Simulation-Based Optimization

Expensive computer simulations pose interesting challenges in design optimization. Conservation of function evaluations and the ability to deal with noisy responses become prime concerns. A common approach is to create an inexpensive approximation that may also smooth out some of the small scale noise. Typically, an approximation is made for the objective function and each of the constraint functions, which is ine cient if some of the functions (either objective or constraint) are not expensive. This paper presents modi cations to the surrogate model-based optimization algorithm, EGO. By incorporating information from inexpensive functions, analytical tests required between 10% and 50% fewer iterations and 20% to 65% less time. A simulation-based case study showed greater bene ts.