Performance Study of Mode-Pursuing Sampling Method

Since the publication of our recently developed mode-pursing sampling (MPS) method, questions have been asked on its performance as compared with traditional global optimization methods such as genetic algorithm (GA), and when to use MPS as opposed to GA. This work aims to provide an answer to these questions. Similarities and distinctions between MPS and GA are presented. Then MPS and GA are compared via testing with benchmark functions and practical engineering design problems. These problems can be categorized from different perspectives such as dimensionality, continuous / discrete variables, or the amount of computational time for evaluating the objective function. It is found that both MPS and GA demonstrate great effectiveness in identifying the global optimum. In general, MPS needs much less function evaluations and iterations than GA, which makes MPS suitable for expensive functions. But GA is more efficient than MPS for inexpensive functions. In addition, MPS is limited by the computer memory when the total number of sample points reaches a certain extent. This work serves a purpose of positioning the new MPS in the context of direct optimization and provides guidelines for users of MPS. It is also anticipated that the similarities in concepts, distinctions in philosophy and methodology, and effectiveness as direct search methods for both MPS and GA will inspire the development of new direct optimization methods.