Dynamic differential evolution with combined variants and a repair method to solve dynamic constrained optimization problems: an empirical study.

An empirical study of the algorithm dynamic differential evolution with combined variants with a repair method (DDECV $$+$$ Repair) in the solution of dynamic constrained optimization problems is presented. Unexplored aspects of the algorithm are of particular interest in this work: (1) the role of each one of its elements, (2) its sensitivity to different change frequencies and change severities in the objective function and the constraints, (3) its ability to detect a change and recover after it, besides its diversity handling (percentage of feasible and infeasible solutions) during the search, and (4) its performance with dynamism present in different parts of the problem. Seven performance measures, eighteen recently proposed test problems and eight algorithms found in the specialized literature are considered in four experiments. The statistically validated results indicate that DDECV $$+$$ Repair is robust to change frequency and severity variations, and that it is particularly fast to recover after a change in the environment, but highly depends on its repair method and its memory population to provide competitive results. DDECV $$+$$ Repair shows evidence on the convenience of keeping a proportion of infeasible solutions in the population when solving dynamic constrained optimization problems. Finally, DDECV $$+$$ Repair is highly competitive particularly when dynamism is present in both, objective function and constraints. [ABSTRACT FROM AUTHOR]