Recently, there have been developed several promising multiobjective evolutionary algorithms such as SPEA, PAES, NSGA-II, PESA, and SPEA2. In those multiobjective evolutionary algorithms, much effort has been concentrated on the elitism and selection technique of the parent and archive solutions. Recently, most of the multiobjective evolutionary algorithms adopt the elitism to improve the efficiency and effectiveness of the multiobjective evolutionary algorithms by utilizing repeatedly the solutions with good characteristics (objective values). And also parent and archive solutions are selected based on the Pareto rank and distribution density so that the best coverage and uniformity of found solutions can be achieved. However, all of the evolutionary algorithms use the same offspring generation scheme such as crossover or mutation operator and new offspring generation schemes have not been tried.
In this thesis, I will propose new three kinds of multiobjective evolutionary algorithms to improve the efficiency and the effectiveness of the multobjective evolutionary algorithm by introducing new offspring generation schemes. In all of the multiobjective algorithms proposed in this thesis, the elitism and selection technique mentioned above are utilized with no change.
In the first offspring generation scheme, special offspring generation structure is used to extract the Pareto dominance information among the solutions and introduce an important concept, “age”, to determine the evolution distance of the solutions. Age concept is an important concept in three kinds of multiobjective evolutionary algorithms proposed in this thesis so that it will be repeatedly and widely utilized because the convergence level can be estimated to a certain extent by evaluating the age of solutions. However, this offspring generation scheme cannot be applied to the problem with high parameter dimensions due to the characteristics of the offspring generation scheme.
To cope with t...