Metamodels in engineering optimization

Modern systems engineering often relies on resource-intensive computer simulations which approximate real-world physics, and this in turn casts the design process as an optimization problem. Metamodels are often used in this context to approximate the response of the computer simulations and to thereby significantly reduce the computational resources needed. A variety of metamodels have been proposed with each having its specific characteristics and suitability to different scenarios. Furthermore, using a metamodel variant which is inadequate to the problem being solved can degrade the effectiveness of the optimization process and can result in a poor final solution. Accordingly it is necessary to select a suitable metamodel, and to facilitate in this aspect this study surveys several mainstream metamodel types and compares their features. It also discusses more advanced approaches which enhance the performance of a metamodel-assisted optimization search. Overall, the study provides a review and discussion along with practical guidelines on the application of metamodels in real-world optimization problems.