Comparing the N-Branch Genetic Algorithm and the Multi-Objective Genetic Algorithm

摘要

An investigation using four test cases compared performance of a selection-based and a ranking fitness-based GA approach on different types of three-objective problems. For these problems, both approaches reasonably approximate the Pareto set. Results of the simple constrained and 10-bar truss problems suggest that the N-branch GA is better suited to solving constrained problems. The adjacent minima problem results favor the MOGA approach. Golinski's speed reducer is both highly constrained and has two similar objectives, and it is not clear which approach is preferred here. The MOGA finds more evenly distributed points on the Pareto front than the N-branch GA; however, the N-branch GA finds better solutions near the extremes of the Pareto set where constraints are active. Based on the investigations, it is difficult to make a definitive choice for the preferred MOGA approach. Like many aspects of the GA, the choice appears problem dependent. It seems that the N-branch tournament GA is preferred for three-objective problems with solutions that have numerous active conslraints. The MOGA appears preferred for problems in which two or more objectives are adjacent in the design space. However, for many engineering problems, it may not be known whether two or more objectives are adjacent a priori. The success of the N-branch tournament GA demonstrates that a selection-based method can generate multi-objective solutions that are comparable to those generated by a nondominance ranking method.