A Two-Stage Algorithm for Integer Multiobjective Simulation Optimization

Fei Liu, Qingfu Zhang

Abstract: Multiobjective discrete optimization via simulation (MDOvS) has received considerable attention from both academics and industry due to its wide application. This paper proposes a two-stage fast convergent search algorithm for MDOvS. In its first stage, the multiobjective optimization problem under consideration is decomposed into several single-objective optimization subproblems, and a Pareto retrospective approximation method is used to generate an approximated optimal solution for each subproblem. In the second stage, from the solutions generated in the first stage, a multiobjective local stochastic search with a revised simulation allocation rule is used to explore the entire Pareto front. Our experimental studies show that the proposed method outperforms the state-of-the-art MO-COMPASS on a set of test instances with noisy evaluations and a bi-objective bus scheduling problem. Our proposed method is up to ten times faster than MO-COMPASS.