An Algorithm for Expensive Optimization Problems

Computer simulations are used extensively in engineering and science to evaluate candidate designs, as a partial substitute for real-world experiments. Metamodels, which are computationally cheaper approximations of the simulation, are often used in these settings to alleviate various issues arising in such simulation-driven design processes. However, due to the high computational cost of running the simulation only a small number of designs can be evaluated, and hence the resultant metamodel will be inaccurate. To achieve a more accurate approximation, ensembles employ multiple metamodel variants concurrently, and aggregate their individual predictions into a single one. Nevertheless, the optimal ensemble topology, namely, which types of metamodels should be incorporated, is typically not known a-priori, while using a fixed topology may degrade the search effectiveness. To address this issue, this study proposes a new metamodel-assisted algorithm with dynamic topology adaptation, namely, which autonomously adapts the ensemble topology during the search, and dynamically selects the most suitable topology as the search progresses. An extensive performance analysis shows the effectiveness of the proposed algorithm, and highlights the merit of the proposed topology adaptation.