Efficiency speed-up strategies for evolutionary computation: fundamentals and fast-GAs

作者:

Highlights:

摘要

Through identifying the main causes of low efficiency of the currently known evolutionary algorithms, a set of six efficiency speed-up strategies are suggested, analyzed, and partially explored, including those of the splicing/decomposable representation scheme, the exclusion-based selection operators, the “few-generation-ended” EC search, the “low-resolution” computation with reinitialization, and the coevolution-like decomposition. Incorporation of the strategies with any known evolutionary algorithm leads to an accelerated version of the algorithm. On the basis of problem space discretization, the proposed strategies accelerate evolutionary computation via a “best-so-far solution” guided, exclusion-based space-shrinking scheme. With this scheme, an arbitrarily high-precision (resolution) solution of a high-dimensional problem can be obtained by means of a successive low-resolution search in low-dimensional search spaces. As a case study, the developed strategies have been endowed with genetic algorithms (GAs), yielding an accelerated genetic evolutionary algorithm: the fast GAs. The fast-GAs are experimentally tested with a test suit containing 10 complex multimodal function optimization problems and a difficult real-life problem––the moment matching problem for inverse to the fractal encoding of the spleenwort fern. The performance of the fast-GA is compared against the standard GA (SGA) and the forking GA (FGA) (that is one of the most recent and fairly established variants of GAs). The experiments all demonstrate that the fast-GAs consistently and significantly outperform the SGA and FGA in efficiency and solution quality in the test cases. Besides the speed-up of efficiency, other visible features of the fast-GAs include: (i) no premature convergence occurs; (ii) better convergence capability to global optimum; and (iii) variable high-precision solutions attainable. All these support the validity and usefulness of the developed strategies.

论文关键词:Evolutionary computation,Genetic algorithms,Efficiency speed-up,Space discretization,Refined cellular partioning,Space shrinking,Splicing/decomposable representation scheme,Exclusion-based selection operator,Reinitialization,Coevolution-like decomposition,Fractal image compression

论文评审过程:Available online 7 January 2003.

论文官网地址:https://doi.org/10.1016/S0096-3003(02)00309-0