Multiobjective optimization design via genetic algorithm

Many real-world problems involve multiple objectives that need to be optimized simultaneously. However, in most cases, a suitable optimal solution meeting all the objectives can hardly be found since these objectives are generally conflicting. Compared to conventional optimization techniques, genetic algorithms (GAs) are well suited to solve multiobjective optimization (MO) problems since a family of "acceptable" solutions-a so called Pareto set-can be identified by different individuals through the evolution process. However, most of the existing multiobjective optimization genetic algorithms (MOGAs) have difficulty dealing with the trade-off between uniformly distributing the computational resources and avoiding the "genetic drift" phenomenon. The paper proposes a new evolutionary approach to MO problems-the rank-density based genetic algorithm (RDGA). From the result of the simulation study, RDGA clearly outperforms two representative MOGAs on three benchmark testing problems in terms of keeping the diversity of the individuals along trade-off surface, tending to extend the Pareto front to new areas, and finding a well-approximated Pareto optimal set