Stochastic search methodologies for multi-objective simulation optimization

Abstract

For multi-objective simulation optimization problem (MSOP), this thesis proposed a generic framework for designing various types of multiobjective (MO) search algorithms. Based on the framework, two specific algorithms are designed. First is an advanced stochastic search algorithm multi-objective convergent optimization via most-promising-area stochastic search (MO-COMPASS) that is developed with solid theoretical foundations and proof showing convergence to the local optimum. Another is gradient-oriented polar random search (GO-POLARS) that is designed to strengthen the search efficiency, especially to make it suitable for continuous problems and easy to control exploration of the search space. With the dominated hyper volume concept and a unified gradient derived, we are able to incorporate gradient-based techniques such as GO-POLARS, into the MO search framework.