Adaptive Metamodeling-Based Simulation Optimization

A pillar of Industry 4.0, simulation is a powerful tool used across several fields, enabling system evaluation under varying conditions, facilitating performance analysis, and more efficient decision-making. On the other hand, simulation might be time-consuming, particularly when considering complex model optimization. In this sense, metamodeling has emerged as a popular technique for simulation optimization. This paper presents a novel metamodeling framework called Adaptive Metamodeling-based Simulation Optimization that aims to achieve better solutions using fewer experiments. Our approach combines machine learning and metaheuristic techniques to identify the most promising areas of the solution space that can be explored more efficiently to achieve optimal solutions. The proposed framework is evaluated in two real-world problems, a resource allocation of a manufacturing digital twin model and a mining expansion project. Compared to the Efficient Global Optimization method, the AMSO found a solution 8.1% better (on average) in the first case and 9.7% in the second case. Moreover, AMSO found solutions that were statistically equal to the Genetic Algorithm method but required 5.5 and 10.7 less computational time in the first and second cases, respectively