An Unconstrained Auxiliary Framework for Constrained Many-Objective Optimization

Constrained many-objective optimization problems (CMaOPs) include the optimization of many objective functions and satisfaction of constraints, which seriously enhance the difficulty of problems. Although several constrained many-objective evolutionary algorithms (CMaOEAs) have been designed, they still have difficulties in tackling many objectives and constraints at the same time. To better solve CMaOPs, this paper proposes an unconstrained auxiliary framework, in which an auxiliary task without constraints is developed to reduce the search difficulties of constraints. Moreover, to tackle many objectives, the existing CMaOEAs are employed to address the auxiliary task, in which the constraint values of solutions are set to zeros. In the experiments, one classic CMaOEA and two latest CMaOEAs are integrated into the framework to form three new algorithms. The results show the effectiveness and superiority of the framework. Besides, the winner among three new algorithms is compared with several existing CMaOEAs and shows better results. Meanwhile, we discuss the reasons that why the unconstrained framework is effect for the existing benchmark functions. Accordingly, we refer to that new test functions are urgently needed for the development of constrained many-objective optimization.