ON SOLVING MULTI-OBJECTIVE SIMULATION OPTIMIZATION BY OPTIMAL COMPUTING BUDGET ALLOCATION AND RANDOM SEARCH

Abstract

For decision-making in large-scale complex stochastic systems considering multiple conflicting objectives, analytical models and closed-form optimal solutions are usually hard to formulate and derive. These issues can be addressed by multi-objective simulation optimization, which employs efficient simulation to evaluate solutions' performance and such information is further used to guide the optimization. Despite the advantages of simulation optimization, there exist many challenges to be solved. In this thesis, three approaches are proposed to address these challenges through optimal computing budget allocation and random search. First, three simulation budget allocation strategies with asymptotic optimal guarantee are developed to tackle with multi-objective Ranking and Selection from the perspective of large deviation principle. Second, two globally convergent partition-based random search algorithms are developed to solve multi-objective optimization via simulation. Last, a less conservative approach is proposed to attack robust Ranking and Selection with input uncertainty.