A Comparison of Multi-objective Evolutionary Algorithms for Simulation-Based Optimization

Abstract

Simulation-based optimization is an important tool in science, engineering, business and many other areas. Optimization of a real-world physical system often involves multiple (and sometimes conflicting) objectives. This gives rise to a situation where a set of optimal solutions, also known as the Pareto-optimal front (POF), is applicable. For non-trivial problems, the number of possible solutions is typically very large. This makes it impossible to apply an exhaustive search to find all possible solutions in the POF in a reasonable time. By applying heuristic search algorithms, such as evolutionary algorithms, it is possible to search for Pareto-optimal solutions without having to evaluate the entire search space. While heuristics can help to reduce the number of solutions that need to be evaluated, there is still the issue of having to perform multiple simulation replications due to the stochastic nature of many simulation models. Since simulation is time-consuming, it is important to implement a computing budget allocation scheme to ensure the simulation is completed within a reasonable time. The research presented in this paper examines the impact of dynamic computing budget allocation on the performance of evolutionary algorithms with respect to the quality of solutions. The results show that the use of dynamic computing budget allocation in combination with an integrated evolutionary algorithm has comparable performance while using less computing budget when compared to a standard approach.