Investigating the use of sequencing and infeasibility driven strategies for constrained optimization

Real-world optimization problems involve constraints that must be satisfied for the design to be viable. The constraints are a manifestation of statutory physical limitations such as allowable strength, geometric compatibility or other practical considerations such as cost and time required for manufacturing. Constraint handling is thus an important area in the domain of optimization and there exists rich literature on the subject. Within population based stochastic optimization methods, constraint handling is typically implemented through a ranking process, where feasible solutions are considered better then infeasible ones. Recent studies have suggested that preserving infeasible solutions can be advantageous to the evolutionary search. Such studies have typically considered a paradigm where all objectives and constraints are evaluated simultaneously. In many practical scenarios however, it is possible to evaluate them independently. This opens up opportunities to gain computational benefits through sequencing constraint evaluation and using partial evaluation (i.e. only evaluate some of the constraints). In this paper, we systematically construct and study the performance of these strategies by combining them in different ways (total 8 variants). The numerical experiments compare the performance of these strategies under different evaluation and ranking scenarios. The study offers understanding of the advantages that can be gained by using appropriate combinations of these strategies for the cases where objective and constraint(s) are associated with individual cost and can be computed independently.