Multiscale multiresolution genetic algorithm with a golden sectioned population composition

Abstract

A new genetic algorithm (GA) strategy called the multiscale multiresolution GA is proposed for expediting solution convergence by orders of magnitude. The motivation for this development was to apply GAs to a certain class of large optimization problems, which are otherwise nearly impossible to solve. For the algorithm, standard binary design variables are binary wavelet transformed to multiscale design variables. By working with the multiscale variables, evolution can proceed in multiresolution; converged solutions at a low resolution are reused as a part of individuals of the initial population for the next resolution evolution. It is shown that the best solution convergence can be achieved if three initial population groups having different fitness levels are mixed at the golden section ratio. An analogy between cell division and the proposed multiscale multiresolution strategy is made. The specific applications of the developed method are made in topology optimization problems.The optimal choice for the population size increase is still a subject of further studies, the overall GA performance will not be significantly affected if the size increase is not too small.

Establishing the Optimal Density of the Michell Truss Members

“…Selection technology allows iteration acceleration while maintaining the decision sensitivity of evolutionary algorithms. Publications [ 12 , 13 , 14 , 15 , 16 ] deal with this in more detail.…”

Automated Design for Microfluid Flow Sculpting: Multiresolution Approaches, Efficient Encoding, and CUDA Implementation

“…Thus, a single GA is modified during evolution as in Algorithm 2. This is an example of the so-called "multiscale" GA, which has been used to improve the rate of convergence [17,18]. We recognize that the choice of generation switch and resolution is a complete study unto itself, and defer an exhaustive study to a future publication.…”

Computational methods and software for the design of inertial microfluidic flow sculpting devices

“…Thus, a single GA is modified during evolution as in Algorithm 2. This is an example of the so-called "multi-scale" GA, which has been used to improve the rate of convergence (Kim et al, 2008;Babbar and Minsker, 2006). We recognize that the choice of generation switch and resolution is a complete study unto itself, and defer an exhaustive study to a future publication.…”