Explicit Building-Block Multiobjective Genetic Algorithms: Theory, Analysis, and Development.

摘要

This dissertation research emphasizes explicit Building Block (BB) based MO EAs performance and detailed symbolic representation. An explicit BB- based MOEA for solving constrained and real-world MOPs is developed the Multiobjective Messy Genetic Algorithm II (MOMGA-II) which is designed to validate symbolic BB concepts. The MOMGA-II demonstrates that explicit BB-based MOEAs provide insight into solving difficult MOPs that is generally not realized through the use of implicit BB-based MOEA approaches. This insight is necessary to increase the effectiveness of all MOEA approaches. In order to increase MOEA computational efficiency parallelization of MOEAs is addressed. Communications between processors in a parallel MOEA implementation is extremely important, hence innovative migration and replacement schemes for use in parallel MOEAs are detailed and tested. These parallel concepts support the development of the first explicit BB-based parallel MOEA the pMOMGA-II. MOEA theory is also advanced through the derivation of the first MOEA population sizing theory. The multiobjective population sizing theory presented derives the MOEA population size necessary in order to achieve good results within a specified level of confidence. Just as in the single objective approach the MOEA population sizing theory presents a very conservative sizing estimate. Validated results illustrate insight into building block phenomena good efficiency excellent effectiveness and motivation for future research in the area of explicit BB- based MOEAs. Thus the generic results of this research effort have applicability that aid in solving many different MOPs.