Adaptively preserving solutions in both feasible and infeasible regions on generalized multiple constraint ranking
Yohanes Bimo Dwianto, Hiroaki Fukumoto, Akira Oyama · 2020
In the present work, we propose some new modifications of an existing constraint handling technique (CHT) for single-objective optimization problems. The base CHT is generalized multiple constraint ranking (G-MCR), which is already a modified version of the original CHT, MCR. Despite that G-MCR significantly outperformed the original MCR in the previous study, it is found that G-MCR tends to generate very few feasible individuals on a certain real-world like engineering design problem. In the present work, G-MCR is further modified to strike a better balance between feasible and infeasible individuals on each generation in an adaptive way so that the interaction between feasible and infeasible regions can be maintained, thus providing more efficient search towards constrained global optimum. Based on the investigation on 78 benchmark problems, we obtain that some of the proposed modifications produce more robust convergence performance by obtaining significant superiority on many types of problems. On real-world like engineering design problems, we also observe that the feasibility ratio generated on each generation might have an important role in improving the convergence performance.