Multiobjective genetic algorithms made easy: selection sharing and mating restriction

摘要: This paper aims to illustrate how an existing GA can be modified and set up explore the relevant trade-offs between multiple objectives with a minimum of effort. While Pareto Pareto-like ranking schemes easily implemented, current guidelines on associated set-up techniques such as sharing mating restriction are intricate and/or based more or less rough assumptions about cost landscape, making them impractical. However, if fitness is reinterpreted technique involving estimation population density at points defined by each individual so-called kernel methods, setting parameter comes depend only size distribution population, not problem. Kernel estimation, from statistics data analysis, introduced shown find direct application in restriction, simplifying implementation avoiding introduction any tunable parameters formulation. After brief multiobjective optimization discussion preference articulation GAs, main differences single-objective GAs highlighted, conversion into described means example. Simple experimental results presented towards end paper.