A Tunable Constrained Test Problems Generator for Multi-objective Optimization

Abstract

Multi-objective optimization problems (MOPs) in real world are often constrained optimization problems. So test problems to evaluate multi-objective optimization evolutionary algorithms (MOEAs) should have some constraints in order to simulate real-world problems. In this paper, a well understood and tunable constrained test problems generator is suggested. By setting parameters in the constraint function, test problems with various complexity and Pareto-optimal front geometries can be created. Six constrained MOPs are developed and explained in figures so as to account for parameters in the constraint function. Furthermore, NSGA-II with an constrained handling strategy are used to solve the test problems. Experiments results show test problem can greatly increase difficulties in searching Pareto-optimal solutions, and they are effective tools to evaluate MOEAs in constraints handling.